Small Biz AI HQ

Author: Small Biz AI HQ Editorial Team

  • AI in Real Estate: Valuation, Search, and the End of the Endless Scroll

    AI in Real Estate: Valuation, Search, and the End of the Endless Scroll

    Buying, selling, and renting property has long involved endless scrolling through listings, waiting for valuations, and paperwork. AI is changing each step. Automated valuation models estimate prices in seconds, property portals are experimenting with search you can talk to, and agents use AI to write listings, stage empty rooms virtually, and answer enquiries around the clock. This briefing covers the main uses, the risks, and what smaller agencies, landlords, and property businesses can do.

    Automated valuations

    Automated valuation models (AVMs) estimate a property’s value using sold prices, property characteristics, location data, and market trends. Lenders use them for some mortgage decisions and remortgages, and portals show estimated values to homeowners.

    AVMs are fast and cheap, and work best for standard homes in areas with many comparable sales. They struggle with unusual properties, rural homes, recent renovations, and fast-moving markets, because they can’t see condition, light, noise, or a new kitchen. For agents, they’re a useful starting point for a valuation, not a replacement for local knowledge and a visit.

    Search that understands what people want

    Traditional property search relies on filters: price, bedrooms, location. AI-powered search lets people describe what they want in plain language: “a quiet three-bed with a garden, within 20 minutes’ walk of a good primary school and a station.” Systems can also analyse listing photos to find features such as open-plan kitchens or wooden floors that aren’t in the text. The promised result is less scrolling and better matches, though these features are still developing.

    Listing content

    Agents use AI to:

    • draft property descriptions from notes and photos;
    • write social media posts and email alerts;
    • create floorplans from phone scans;
    • produce virtual staging — furnishing empty rooms digitally;
    • enhance photos (lighting, sky replacement).

    Accuracy matters legally. In the UK, agents must not give misleading information about properties under consumer protection rules, and National Trading Standards has guidance on material information that listings must include. Virtually staged images should be clearly labelled, and edits must not hide defects or misrepresent size, views, or condition.

    Lead handling and customer service

    AI assistants can answer common questions, book viewings, qualify leads, and follow up, including outside office hours when many buyers and tenants browse. The best setups hand over quickly to a person for negotiation and sensitive conversations.

    Lettings and property management

    For landlords and letting agents, AI helps with:

    • tenant enquiry handling and viewing scheduling;
    • maintenance request triage, with photos analysed to identify the issue and route it to the right contractor;
    • rent arrears prediction and gentle reminders;
    • document processing for tenancy agreements and compliance certificates.

    Compliance deadlines — gas safety, electrical checks, EPCs, deposit protection — can be tracked automatically, but responsibility remains with the landlord or agent.

    Tenant screening and fairness

    Some screening tools use AI to assess applicants. This raises fairness concerns: models can disadvantage people based on proxies for protected characteristics, and automated decisions have legal implications under UK GDPR and equality law. Decisions should be explainable, and applicants should be able to challenge them. Keep human review in the process.

    Market analysis and investment

    Investors and developers use AI to analyse planning data, price trends, rental yields, and local demand. These tools can speed up research, but forecasts rely on past patterns and can be wrong when markets shift.

    Risks summary

    • Inaccurate valuations for unusual properties.
    • Misleading AI-generated or edited listing images.
    • Bias in tenant screening.
    • Data protection issues with applicant and tenant data.
    • Generic listings that all sound the same.

    Where a small agency or landlord can start

    1. Use AI to draft listing descriptions, then add local knowledge and check every fact.
    2. Set up automated enquiry responses and viewing booking, with fast human follow-up.
    3. Label virtual staging clearly.
    4. Automate compliance reminders for certificates and renewals.
    5. Use AVMs as a cross-check, not a substitute for your valuation expertise.

    Frequently asked questions

    Are online house valuations accurate?

    They’re reasonable for typical homes in active markets, but can be well off for unusual or improved properties. A local agent or surveyor gives a more reliable figure.

    Is virtual staging allowed?

    Yes, if it’s clearly disclosed and doesn’t misrepresent the property.

    Can AI decide whether to accept a tenant?

    It can support the process, but fully automated decisions carry legal risks. Keep a person responsible for the final decision.

    Related reading

  • How Small Businesses Use Midjourney for Marketing Visuals

    How Small Businesses Use Midjourney for Marketing Visuals

    Good visuals sell, but professional photography and illustration aren’t cheap. Midjourney is an AI image generator known for striking, high-quality images from a text description. For small businesses it’s useful for blog headers, social media backgrounds, mood boards, ad concepts, and seasonal campaign art — as long as you use it for the right jobs and understand the rules.

    What Midjourney is good (and bad) at

    Good at: atmospheric scenes, stylised illustrations, textures and backgrounds, concept art, lifestyle-style imagery, and exploring ideas quickly before paying for a shoot.

    Weak at: accurate depictions of your real products, readable text inside images (improving, but still unreliable), exact brand logos, and anything that must be factually correct — such as your actual premises, staff, or food.

    The rule of thumb: use AI images for mood, use real photos for proof.

    Getting started

    1. Midjourney is a paid service; check midjourney.com for current plans. Higher plans include more generation time and a “stealth” option to keep images private.
    2. Create images in the web app by typing a description (a “prompt”). Each prompt produces a grid of options you can upscale or vary.
    3. Your images appear in a gallery you can organise and download.

    How to write a good prompt

    A useful structure is: subject + setting + style + lighting + composition + format. For example:

    A flat lay of fresh sourdough loaves on a rustic wooden table, morning window light, warm neutral colours, minimal styling, overhead shot, lots of empty space on the right for text

    Tips:

    • Be concrete. “Warm afternoon light in a small independent bookshop” beats “cosy vibe.”
    • Leave space for text if the image will be a banner or ad.
    • Set the shape with the aspect-ratio parameter, for example –ar 16:9 for blog headers or –ar 4:5 for Instagram posts.
    • Iterate. Use the vary and remix options on the closest result rather than starting from scratch.

    Keeping a consistent brand style

    Random styles from post to post look amateurish. To stay consistent:

    • write a short style phrase you add to every prompt, such as “soft pastel palette, clean flat illustration, lots of white space”;
    • use Midjourney’s style reference feature with an image you like, so new images follow its look;
    • save your best prompts in a shared document;
    • finish images in Canva with your brand fonts, colours, and logo.

    Practical uses for a small business

    • Blog and newsletter headers — illustrative images that suit the topic.
    • Social media backgrounds for quotes, tips, and announcements.
    • Seasonal campaigns — autumn, Christmas, summer sale artwork.
    • Ad concepts — test several visual ideas before investing in a photoshoot.
    • Mood boards — show a designer, photographer, or shopfitter the look you want.
    • Presentation visuals for pitches and proposals.

    Commercial use and legal points

    • Check the terms. Midjourney’s terms explain who owns the images and what’s allowed commercially, including rules that depend on your plan and business size. Read the current version.
    • Copyright protection is limited. In several countries, purely AI-generated images may not be protected by copyright in the same way as human-made work, so competitors might be able to reuse them. Don’t build your core brand identity — especially your logo — purely from AI output.
    • Avoid real people and brands. Don’t generate images of real people, celebrities, or other companies’ logos and characters, or copy a living artist’s distinctive style for commercial work.
    • Don’t mislead customers. Showing a dish, room, or product that doesn’t match reality can break advertising rules. The UK Advertising Standards Authority expects ads to be accurate whatever tools made them.
    • Disclose where required. Some ad platforms require you to label realistic AI-generated images.

    A simple workflow

    1. Write the message first (the headline or offer).
    2. Generate four to eight image options with your style phrase.
    3. Pick one, upscale it, and download.
    4. Add text and logo in Canva, then resize for each channel.
    5. Keep a record of the prompt in case you need a matching image later.

    Frequently asked questions

    Is there a free version of Midjourney?

    Midjourney is generally a paid tool. For occasional images, free tiers from other generators, including Canva’s text-to-image tool, may be enough.

    Can I use Midjourney images in paid ads?

    Generally yes, within Midjourney’s terms and the ad platform’s policies. Make sure the image doesn’t mislead and doesn’t include protected brands or real people.

    Should I use AI images for my product pages?

    Use real, accurate product photos. AI can help with backgrounds or lifestyle scenes, but customers need to see what they’re actually buying.

    Related reading

  • How Small Businesses Use Zapier’s AI Automations

    How Small Businesses Use Zapier’s AI Automations

    Most small businesses lose hours every week to copy-and-paste work: moving enquiries from a website form into a spreadsheet, sending the same welcome email, updating the CRM after every call, chasing invoices. Zapier connects thousands of apps so that when something happens in one, an action happens in another. Its AI features make those automations smarter — reading an email to decide what it’s about, summarising a form response, or drafting a reply for you to check.

    How Zapier works in plain English

    A Zapier automation is called a Zap. Each Zap has:

    • a trigger — the event that starts it, such as “new form submission” or “new email with a label”;
    • one or more actions — what happens next, such as “add a row to Google Sheets” or “send a Slack message”;
    • optional filters and paths — rules like “only continue if the enquiry is over £1,000.”

    AI steps sit in the middle. They can classify, summarise, extract details, or draft text using an AI model, and pass the result to the next step. Zapier also lets you describe an automation in plain English and builds a draft Zap for you to review.

    Getting started

    1. Sign up at zapier.com. The free plan covers simple two-step Zaps; paid plans add multi-step Zaps, filters, and more tasks per month. Check current pricing, since plans change.
    2. List your repetitive jobs for one week. Anything done more than a few times a week with the same steps is a candidate.
    3. Start with one small automation, test it, and only then build the next.

    Five automations worth building first

    1. Enquiry triage

    Trigger: new website form entry. AI step: classify the enquiry (quote request, complaint, job application, spam) and summarise it in one line. Actions: add it to a spreadsheet or CRM, and send urgent ones to your phone or team chat. Result: nothing gets lost in an inbox, and the important messages reach you first.

    2. Draft replies for approval

    Trigger: new email in a shared inbox with a specific label. AI step: draft a reply using your FAQ text. Action: save the draft in your email account — not send it. A person reviews and sends. This keeps quality high and avoids embarrassing automatic replies.

    3. Meeting notes to action points

    Trigger: a new transcript from your meeting tool. AI step: extract decisions, actions, and owners. Action: create tasks in your project tool and email a summary to attendees.

    4. Review monitoring

    Trigger: new review on a supported platform. AI step: detect sentiment and topic. Actions: log it in a sheet and alert you immediately to negative reviews so you can respond quickly.

    5. Invoice and payment follow-up

    Trigger: an invoice in your accounting software becomes overdue. Action: send a polite, personalised reminder; after a second reminder, notify you to call the customer. Keep the tone friendly and make sure the details are right before switching it on.

    Testing before you trust it

    • Use test data first. Zapier lets you test each step. Check the output of AI steps carefully.
    • Start with “draft” or “notify” actions, not automatic sending or deleting.
    • Add filters to stop runaway automations — for example, only process emails from outside your domain.
    • Check the Zap history weekly for errors in the first month.
    • Name Zaps clearly so anyone on the team knows what each one does.

    Costs to keep an eye on

    Zapier charges based on tasks (each action that runs). A busy Zap with several steps can use tasks quickly. Watch usage on the dashboard, remove Zaps you no longer need, and use filters early in the Zap so unnecessary runs stop before using many tasks.

    Data protection

    Automations move personal data between services. Under UK GDPR you should know where customer data goes, only send what’s needed, and make sure each connected service is covered in your privacy policy. Avoid sending sensitive data to AI steps unless you’re comfortable with the provider’s data terms.

    Common mistakes

    • Automating a messy process instead of fixing it first.
    • Letting AI send emails to customers with no human check.
    • Building one huge Zap instead of several simple ones.
    • Forgetting that apps change: a renamed form field can quietly break a Zap.

    Frequently asked questions

    Do I need coding skills for Zapier?

    No. Zaps are built with menus and plain-English descriptions. Basic logic — “if this, then that” — is all you need.

    Zapier or Make?

    Both connect apps. Zapier is usually simpler to learn; Make offers more visual control and can be cheaper for complex workflows. Try your first automation in each and compare.

    Can Zapier use ChatGPT?

    Yes. You can connect ChatGPT and other AI services, or use Zapier’s own AI steps. See our ChatGPT guide for writing good prompts.

    Related reading

  • How Small Businesses Use Perplexity for Research

    How Small Businesses Use Perplexity for Research

    Research is one of those jobs small business owners know they should do more of — checking competitors, finding suppliers, understanding a regulation, sizing up a new market — but rarely have time for. Perplexity is an AI search tool built for exactly this. Instead of ten blue links, it gives you a written answer with numbered citations you can click to check.

    That last part matters. The value of Perplexity isn’t that it’s always right; it’s that it shows you where each claim came from, so you can verify quickly.

    How Perplexity differs from ChatGPT and Google

    • Versus Google: Google gives you pages to read. Perplexity reads several sources and writes a summary, with links. It’s faster for questions that need information pulled together from several places.
    • Versus a general chatbot: Perplexity searches the live web by default and cites sources. General assistants may answer from training data unless their search feature is used.
    • Follow-up questions: you can keep asking in the same thread, and it keeps the context.

    Getting started

    1. Go to perplexity.ai and create a free account. A paid plan adds more advanced searches, model choices, and file uploads; check current pricing.
    2. Try the focus options where available — for example, limiting results to academic sources or discussions — when you need a particular kind of source.
    3. Use Spaces (or collections) to group research on one topic, such as “New shop location” or “Supplier options.”

    Seven research jobs it handles well

    1. Competitor snapshots

    “Summarise what [competitor] offers, their pricing if public, and what customers say about them in reviews.” You get a starting point in a minute. Then open the sources: review sites and the competitor’s own pages.

    2. Finding suppliers

    “List UK wholesale suppliers of compostable coffee cups that sell to small cafés, with minimum order quantities if listed.” Contact the shortlisted suppliers directly to confirm details — listings go out of date.

    3. Understanding regulations

    “What are the rules for selling homemade food online in England? Link to official guidance.” Perplexity is useful for finding the right official pages — GOV.UK, the Food Standards Agency, HMRC, the ICO — but always read the official source itself, and get professional advice where the stakes are high.

    4. Market and customer research

    “What are the main trends in the UK pet grooming market? Cite industry reports or news articles.” Treat any statistics as leads to verify rather than facts to repeat.

    5. Tool and software comparisons

    “Compare three booking systems for a small hair salon on price, features, and reviews.” Check the vendor pricing pages, because prices change frequently.

    6. Preparing for meetings

    “Summarise recent news about [company] before a sales meeting.” Five minutes of preparation can change the conversation.

    7. Content research

    When writing a blog post or newsletter, use it to gather facts and sources, then write in your own words and link to the original sources.

    How to verify answers properly

    1. Click the citations. Check that the source actually says what the summary claims. Occasionally, a citation is related but doesn’t support the specific point.
    2. Prefer primary sources. Official bodies, company websites, and original research beat blogs quoting other blogs.
    3. Check dates. An article from three years ago may describe rules that have since changed.
    4. Watch for missing context. Summaries compress. For anything important, read the full section.
    5. Cross-check numbers. If a figure matters for a decision, find it in at least two independent sources.

    Prompts that get better results

    • Be specific about location: “in the UK” or “in England” changes the answer on laws and suppliers.
    • Ask for the type of source: “cite official government guidance” or “use sources from the last 12 months.”
    • Ask for structure: “put the answer in a table with columns for name, price, and link.”
    • Ask what it doesn’t know: “what information is uncertain or missing?”

    Limits

    • Summaries can be wrong or oversimplified even with citations.
    • Some high-quality sources are behind paywalls and won’t be included.
    • It isn’t a substitute for legal, tax, or financial advice.
    • Don’t paste confidential business information into queries unless your plan and settings allow it.

    Frequently asked questions

    Is Perplexity free?

    There’s a free tier that handles most everyday research. Paid plans add more advanced search and file features.

    Can I use Perplexity instead of Google?

    Many people use both: Perplexity for questions that need a summary of several sources, and Google for finding a specific website, map, or local business.

    Can I copy Perplexity’s answers into my blog?

    Don’t copy them directly. Use them for research, verify the sources, and write original content in your own words — that’s better for readers and for search rankings.

    Related reading

  • How Small Businesses Use Canva’s Magic Studio

    How Small Businesses Use Canva’s Magic Studio

    Most small businesses don’t have a designer on staff, but they still need social posts, flyers, menus, price lists, and presentations every week. Canva has been the go-to tool for non-designers for years, and its Magic Studio features add AI on top: generating layouts, writing copy, editing photos, and resizing designs for every platform in a click.

    This guide explains the features that save the most time, a simple workflow, and the points to watch when you publish AI-assisted designs.

    What’s in Magic Studio

    Feature names and availability change, and some tools are limited to paid plans, but the core set includes:

    • Magic Design — describe what you need or upload a photo and get a set of ready-made layouts.
    • Magic Write — an AI copywriting assistant inside your designs and documents.
    • Background Remover — cut a product or person out of a photo in one click.
    • Magic Edit and Magic Eraser — replace or remove objects in a photo by brushing over them.
    • Magic Expand — extend a photo beyond its edges to fit a new shape.
    • Magic Resize — convert one design into different sizes, such as an Instagram post, story, Facebook cover, and A4 flyer.
    • Text to image — generate images from a description.

    Step 1: Set up your Brand Kit first

    Before using any AI features, set up a Brand Kit (available on paid plans): logo, brand colours, and fonts. Then apply it to designs so AI-generated layouts come out in your style instead of Canva’s default look. This single step makes the biggest difference to how professional your output looks.

    Step 2: Build a small template library

    Use Magic Design to generate options for your most common formats:

    • a weekly offer post;
    • a new product announcement;
    • an event or opening-hours notice;
    • a customer testimonial (using real reviews only);
    • a flyer or poster.

    Pick the best layout for each, apply your brand kit, and save it as a template. From then on, each new post is a quick edit, not a new design.

    Step 3: Use AI to fix your photos

    Product photos taken on a phone can look professional with a few edits:

    1. Remove the background and place the product on a clean colour or simple scene.
    2. Use Magic Eraser to remove clutter such as cables, price stickers, or passers-by.
    3. Use Magic Expand to turn a portrait photo into a landscape banner.

    Keep product photos accurate. Don’t change colours, sizes, or features in a way that misleads customers — that can break consumer protection rules as well as trust.

    Step 4: Write copy with Magic Write — then edit

    Magic Write can draft captions, flyer text, and slide content. Give it specifics: “Caption for a 20% off weekend sale on garden furniture, friendly, under 40 words, include a call to action to visit the shop.” Always check prices, dates, and terms yourself.

    Step 5: Resize for every channel

    Once a design is finished, Magic Resize creates versions for other platforms. Check each one — text sometimes needs moving so it isn’t hidden behind profile pictures or buttons on stories.

    A 30-minute weekly content routine

    1. 5 minutes: list the week’s messages (offers, news, a tip, a customer story).
    2. 15 minutes: open your saved templates and update text and photos.
    3. 5 minutes: resize for each platform.
    4. 5 minutes: schedule posts with Canva’s Content Planner or your usual scheduling tool.

    Using AI images commercially: what to check

    • Read Canva’s current terms on AI-generated content. They set out what you can do with it and what’s prohibited.
    • Avoid real people, brands, and logos in prompts. Don’t generate images that look like a real person or copy someone else’s trademark or artwork.
    • Don’t mislead. An AI image of a dish you don’t serve, or a “photo” of premises that doesn’t exist, damages trust and may break advertising rules.
    • Label where appropriate. Some platforms require you to disclose realistic AI-generated or altered content, especially in ads.

    Common mistakes

    • Using AI layouts without a brand kit, so every post looks different.
    • Cramming too much text into one graphic.
    • Publishing Magic Write copy without checking prices and dates.
    • Using generated images where a real photo of your product would build more trust.

    Frequently asked questions

    Do I need Canva Pro for Magic Studio?

    Some features are available on the free plan with limits; many, including Brand Kit and Background Remover, need a paid plan. Canva also offers free access for eligible non-profits and education.

    Can I trademark a logo made with Canva’s AI?

    Using stock elements or AI output in a logo can make it hard to protect. For a logo, it’s usually better to work with a designer and own the rights clearly.

    Is Canva better than Midjourney for images?

    They do different jobs. Canva is a complete design tool with AI features; Midjourney focuses on high-quality image generation. Many businesses create images in Midjourney and finish designs in Canva.

    Related reading

  • How Small Businesses Use Claude

    How Small Businesses Use Claude

    Claude, made by Anthropic, is an AI assistant that many small businesses reach for when the job involves long documents and careful writing: reading a supplier contract, summarising a 40-page tender, turning meeting notes into a clear plan, or rewriting a policy so staff actually understand it. It can do general chat too, but its strengths show when you give it a lot of material and ask for a thoughtful answer.

    What Claude is good at

    • Reading long material. You can upload PDFs, spreadsheets, and text files and ask questions about them. It handles long inputs well, so you can work with whole reports rather than pasting pieces.
    • Natural, careful writing. Drafts tend to sound less like marketing copy and follow tone instructions closely.
    • Structured thinking. It is good at pros-and-cons lists, step-by-step plans, and explaining its reasoning.
    • Following detailed instructions. Long, specific briefs usually produce what you asked for.

    Like every AI assistant, it can still be wrong. It may misread a table or state something confidently that isn’t in the document. Always check important points against the source.

    Getting started

    1. Sign up at claude.ai. There is a free plan; paid plans raise usage limits and add features. Check Anthropic’s site for current pricing.
    2. Set your preferences. In settings, describe your business and writing style, such as “UK accountancy practice, formal but plain English, British spelling.”
    3. Use Projects (on paid plans) to keep related work together. Add your price list, brand guidelines, FAQs, and standard terms as project knowledge so every chat starts with the right background.
    4. Agree what data can go in. Remove personal data you don’t need, and check the privacy settings for your plan. Business and team plans offer stronger controls.

    Six practical uses

    1. Contract and terms review (first pass)

    Upload a supplier agreement and ask: “List the renewal terms, notice periods, price increase clauses, liability limits, and anything unusual. Quote the clause numbers.” This gives you a checklist to discuss with your solicitor. It doesn’t replace legal advice.

    2. Tender and bid responses

    Upload the tender document and your company information. Ask Claude to pull out every requirement and scoring criterion into a table, then draft answers section by section. You’ll still need to add evidence and case studies.

    3. Turning notes into plans

    Paste rough meeting notes and ask for decisions, action points with owners, and open questions. It’s a quick way to send a clear follow-up email the same day.

    4. Rewriting for clarity

    “Rewrite this health and safety procedure for new staff whose first language may not be English. Short sentences, numbered steps, no jargon.” It’s a good fit for handbooks and training materials.

    5. Analysing feedback

    Paste anonymised customer reviews or survey answers and ask it to group comments into themes, count how often each appears, and suggest three improvements. Spot-check the grouping against the raw comments.

    6. Customer-facing writing

    Website copy, FAQs, and email sequences, especially when you give it examples of your existing tone.

    Prompts that work well

    • “Here is our returns policy and a customer’s email. Draft a reply that follows the policy exactly. If the policy doesn’t cover the situation, say so rather than guessing.”
    • “Read the attached report. Give me a one-paragraph summary for the board, then five bullet points of risks, each with the page number.”
    • “Act as a sceptical customer reading this sales page. What questions or doubts would you have?”

    Asking it to cite page numbers or quote the source makes checking much faster.

    Limits and risks

    • Accuracy: it can misread numbers in scanned PDFs and complex tables. Verify figures.
    • Current events: its built-in knowledge has a cut-off. Use web search features or official sources for anything recent.
    • Confidential data: follow your plan’s data policy and your own obligations under UK GDPR.
    • Over-reliance: keep a human responsible for anything sent to customers or signed.

    Claude vs ChatGPT

    Both are capable general assistants, and features change often. Many businesses use ChatGPT for quick everyday drafting and Claude for long documents and careful rewrites. The best test is to give both the same real task from your business and compare.

    Frequently asked questions

    Can Claude read Excel files?

    Yes. You can upload spreadsheets and ask for summaries or analysis. For important figures, check the results in the spreadsheet itself.

    Is my data used for training?

    It depends on your plan and settings. Check Anthropic’s current privacy policy and the data controls in your account, and use a business plan for sensitive work.

    Can Claude replace my solicitor or accountant?

    No. It can help you prepare, understand documents, and ask better questions, which can make professional advice faster and cheaper.

    Related reading

  • How Small Businesses Use ChatGPT

    How Small Businesses Use ChatGPT

    If you run a small business, you write the same handful of things over and over: a reply to a customer complaint, a product description, a job advert, a social caption, a policy for the staff handbook. ChatGPT is best used as a first-draft machine for exactly that kind of repeat writing. It won’t know your business, but it is very good at getting you from a blank page to something you can edit in a couple of minutes.

    This guide covers how to set it up properly, the jobs where it saves the most time, prompts you can copy, and the mistakes that cause problems.

    What ChatGPT is (and isn’t)

    ChatGPT is a large language model: it predicts useful text based on patterns in a huge amount of training data. That makes it excellent at structure, tone, and rewording, and unreliable on facts it can’t check. Treat it like a fast, well-read assistant on their first day — helpful, confident, and in need of supervision.

    It is not a source of truth about your prices, your policies, current law, or anything that happened recently unless you give it that information or it searches the web and shows sources.

    Set it up once

    1. Create an account at chat.openai.com. The free tier is enough to test everything in this guide; paid plans add higher limits, newer models, and team features. Check OpenAI’s pricing page for current plans.
    2. Fill in Custom Instructions (Settings → Personalisation). Tell it who you are and how you write, for example: “I run a family-owned bakery in Leeds. Write in British English, friendly but not gushing, short sentences, no emojis unless I ask.”
    3. Decide on a data rule. Don’t paste customer names, card details, health information, or anything confidential into a consumer account. If you need that, look at business plans that exclude your data from model training, and check the settings under Data Controls.
    4. Save your best prompts in a shared document so staff get consistent results.

    Five jobs where it earns its keep

    1. Customer replies

    Paste the customer’s message (with personal details removed) and add: “Draft a polite reply. We can offer a replacement or a refund, not both. Keep it under 120 words.” You get a calm, structured reply in seconds — especially useful for difficult messages you’d otherwise put off.

    2. Product and service descriptions

    Give it the facts: materials, sizes, who it’s for, what makes it different. Ask for three versions of different lengths. Then edit for accuracy. Never let it invent specifications.

    3. Job adverts and interview questions

    “Write a job advert for a part-time sales assistant, 20 hours a week, weekends required, in a garden centre. Include responsibilities, what we offer, and how to apply.” Then ask for ten interview questions based on the advert. Check the wording for anything that could be discriminatory, and confirm pay and legal details yourself.

    4. Policies and procedures

    It is good at turning messy notes into a clear procedure: opening and closing checklists, returns policies, social media guidelines. For anything with legal weight — contracts, employment terms, data protection — use the draft as a starting point for a professional, not a finished document.

    5. Marketing content

    Social captions, email newsletters, blog outlines, and ad variations. Ask for several options and choose. It’s especially useful for repurposing: “Turn this blog post into five LinkedIn posts and one email.”

    Prompt templates you can copy

    • Role + task + constraints: “You are an experienced customer service manager. Draft a reply to this complaint. Constraints: apologise once, offer a 10% discount code, under 100 words.”
    • Give an example: “Here are two product descriptions I like. Write one for [product] in the same style.”
    • Ask it to ask you: “Before you write the newsletter, ask me five questions about what happened in the business this month.”
    • Improve your own draft: “Tighten this email. Keep my voice. Point out anything unclear.”

    The more context you give, the less generic the result.

    Where ChatGPT goes wrong

    • Made-up facts. It can invent statistics, laws, product features, or quotes. Verify anything factual.
    • Generic tone. Without examples it slips into bland marketing language. Give samples of your real writing.
    • Out-of-date knowledge. Prices, regulations, and platform rules change. Check official sources.
    • Overconfidence with numbers. Double-check calculations, especially VAT, margins, and payroll figures.
    • Privacy. What you paste may be stored. Follow your data rule.

    A simple weekly workflow

    1. Monday: ask for a draft of the week’s social posts from your notes.
    2. Daily: use it for tricky customer replies.
    3. Friday: paste the week’s enquiries (anonymised) and ask, “What questions keep coming up? Draft FAQ answers.”

    That last step often produces your most useful website content.

    ChatGPT or another assistant?

    ChatGPT is a strong all-rounder. If you work mostly with long documents, Claude is worth comparing. For research with visible sources, try Perplexity. Many businesses end up using two tools for different jobs.

    Frequently asked questions

    Is ChatGPT free for business use?

    There is a free tier you can use commercially, with limits. Paid plans add capacity, newer models, and — on business plans — stronger data controls. Check OpenAI’s current terms before relying on it.

    Can customers tell the content was written by AI?

    Unedited AI text often sounds generic. Edit every draft, add specific details only you know, and it reads like your business.

    Does Google penalise AI-written content?

    Google’s guidance focuses on whether content is helpful and original, not how it was produced. Thin, mass-produced pages perform poorly whatever made them.

    What should I never paste into ChatGPT?

    Card numbers, passwords, customer personal data, health information, and confidential contracts — unless you’re on a plan and settings approved for that data.

    Related reading

  • AI vs. AI: How Cybersecurity Is Changing on Both Sides of the Fight

    AI vs. AI: How Cybersecurity Is Changing on Both Sides of the Fight

    Cyber security has become a contest of AI against AI. Criminals use it to write convincing phishing emails, clone voices, and scale up attacks. Defenders use it to spot unusual behaviour, filter malicious messages, and respond faster. Small businesses sit in the middle: often targeted because their defences are weaker, and not always aware that the attacks have changed. This briefing explains what’s different and what to do about it.

    How attackers use AI

    Better phishing

    Phishing emails used to give themselves away with poor spelling and awkward phrasing. Generative AI lets criminals write fluent, personalised messages in any language, referencing real suppliers, colleagues, or recent events found online. The UK’s National Cyber Security Centre (NCSC) has assessed that AI will increase the volume and impact of cyber attacks, particularly by making social engineering more effective.

    Deepfake voice and video

    With a short audio sample — from a video, podcast, or voicemail — attackers can clone a voice. There have been reported cases of staff transferring large sums after calls or video meetings with apparently genuine executives. Small businesses face the same trick on a smaller scale: a “director” calling finance with an urgent request.

    Faster reconnaissance and malware

    AI helps attackers research targets, find exposed systems, and adapt malicious code. It lowers the skill barrier, meaning less experienced criminals can launch more capable attacks.

    Attacks on AI itself

    Businesses that use AI tools face new risks: prompt injection (hidden instructions in documents or web pages that manipulate an AI assistant), data leakage through AI tools, and poisoned data.

    How defenders use AI

    • Email filtering that analyses content, sender behaviour, and links to catch phishing that looks legitimate.
    • Endpoint detection and response (EDR) that learns normal behaviour on devices and flags suspicious activity such as ransomware encrypting files.
    • Identity protection that spots impossible logins (for example, the same account signing in from two countries within minutes).
    • Security operations assistants that summarise alerts and suggest responses, helping small IT teams handle more.
    • Fraud detection in banking and payments (see AI in finance).

    Many of these features are built into mainstream business software, such as Microsoft 365 and Google Workspace security tools, so small businesses may already have access without buying new products.

    The small business protection checklist

    Most successful attacks still exploit basic weaknesses. AI makes the basics more important, not less.

    1. Turn on multi-factor authentication (MFA) for email, banking, accounting, and admin accounts. This blocks most password-based attacks.
    2. Use a password manager and unique passwords everywhere.
    3. Keep devices and software updated, including routers, website plugins, and phones.
    4. Back up important data with at least one copy offline or immutable, and test restoring it.
    5. Verify payment and bank detail changes by phone using a known number — never through contact details in the request.
    6. Agree a verification phrase for urgent requests from directors, to defeat voice cloning.
    7. Train staff with short, regular sessions showing real examples of AI-assisted phishing.
    8. Limit admin rights so a compromised account can’t do everything.
    9. Set rules for AI tool use: what data can be pasted in, which tools are approved, and how to treat AI output.
    10. Have an incident plan: who to call, how to isolate systems, and how to report incidents (to Action Fraud, and to the ICO if personal data is involved).

    The NCSC’s Cyber Essentials scheme sets out a solid baseline of technical controls, and certification can help win contracts. The NCSC’s Small Business Guide is free and practical.

    Using AI tools securely

    • Prefer business plans with clear data protection terms.
    • Don’t paste passwords, keys, or sensitive personal data into AI tools.
    • Be careful when AI assistants can take actions — such as sending emails or browsing — because hidden instructions in content can manipulate them.
    • Review AI-generated code or configurations before using them.

    Frequently asked questions

    Are small businesses really targeted?

    Yes. Many attacks are automated and opportunistic, and small businesses often have weaker defences and access to larger customers’ systems.

    Can AI security tools protect me completely?

    No tool offers complete protection. Layered defences — MFA, updates, backups, training, and verification procedures — reduce risk the most.

    What should I do if I’ve been scammed?

    Contact your bank immediately, report it to Action Fraud (in England, Wales, and Northern Ireland) or Police Scotland, change affected passwords, and get IT support to check for compromise.

    Related reading

    *This article is general information, not professional security advice.*

  • AI and the Power Grid: Managing Energy in a More Unpredictable World

    AI and the Power Grid: Managing Energy in a More Unpredictable World

    Electricity grids are being asked to do something they weren’t designed for. They were built around large, predictable power stations. Now they must handle wind and solar output that changes with the weather, millions of electric vehicles and heat pumps, and batteries that both use and supply power. At the same time, data centres — including those that run AI — are adding significant new demand. AI is becoming one of the main tools for managing this complexity, and it’s also part of the problem. This briefing explains both sides and what they mean for businesses.

    Forecasting renewable generation

    Wind and solar output depends on the weather. Grid operators need accurate forecasts hours and days ahead to plan backup generation and avoid waste. Machine-learning models combine weather forecasts, satellite imagery, and historical output data to improve these predictions. Better forecasts mean less expensive standby generation and less need to switch off wind farms when the grid can’t absorb their power.

    Forecasting demand

    Electricity demand changes with temperature, time of day, holidays, and even major TV events. AI improves demand forecasts, including at local level where rooftop solar, EV charging, and heat pumps make patterns less predictable. In Great Britain, the system operator has worked with AI and machine learning to improve forecasting as the grid decarbonises.

    Balancing the grid in real time

    Supply and demand must match every second. AI helps coordinate batteries, flexible demand, and generators to keep frequency stable and prices lower. Demand flexibility is a growing area: businesses and households are paid or given cheaper rates to shift electricity use away from peak times. AI-controlled systems can do this automatically — for example, pre-cooling a cold store before a peak, or charging vehicles overnight.

    Maintaining the network

    Utilities use drones, cameras, and AI to inspect power lines, substations, and wind turbines, detecting damage, corrosion, or vegetation encroachment. Predictive maintenance models analyse equipment data to prioritise repairs before failures cause outages.

    Planning grid connections

    Many renewable and battery projects wait years for grid connections. Network operators are using AI and better data tools to model the grid, speed up connection studies, and find spare capacity. Connection queue reform in Great Britain is also aiming to speed up viable projects.

    AI’s own energy use

    Training and running large AI models requires data centres that use a lot of electricity and, often, water for cooling. The International Energy Agency has projected substantial growth in data-centre electricity demand, driven partly by AI. This raises questions about grid capacity, where data centres are built, and whether new demand is matched by clean power.

    Technology companies are responding with more efficient chips and models, long-term contracts for renewable energy, and in some cases interest in nuclear power. For businesses, it’s a reason to use AI purposefully rather than wastefully — though an individual business’s use of AI tools is small compared with its heating, lighting, and equipment.

    What this means for small businesses

    Small businesses can benefit from the same ideas on a small scale:

    1. Get smart meter data and use your supplier’s or a third-party app’s insights to see when you use energy and what’s running out of hours.
    2. Consider time-of-use tariffs if you can shift use, such as charging vans overnight or running dishwashers, kilns, or compressors off-peak.
    3. Use smart controls for heating, cooling, and lighting that learn occupancy patterns.
    4. Look at solar and batteries with AI-driven controls that decide when to store, use, or export power.
    5. Join a flexibility scheme if you have flexible equipment, such as refrigeration, heating, or EV charging, through your supplier or an aggregator.
    6. Check for grants and support for energy efficiency — schemes vary by region and change often.

    Risks

    • Cyber security: a more connected grid has more targets. Smart devices should be secured and updated.
    • Fairness: customers who can’t shift use or afford smart technology shouldn’t be left paying more.
    • Over-reliance: automated systems need human oversight during unusual events.

    Frequently asked questions

    Does AI make electricity cheaper?

    It can reduce system costs through better forecasting and balancing, but prices depend on many factors, especially wholesale gas prices.

    Is using AI tools bad for the environment?

    Data centres use significant energy overall. For a typical small business, its own AI use is a small share of its footprint; heating, transport, and equipment usually matter far more.

    What’s the easiest energy saving for a small business?

    Understanding your usage from smart meter data and cutting out-of-hours waste is usually the quickest, cheapest win.

    Related reading

  • AI in Transportation and Logistics: Beyond the Self-Driving Headlines

    AI in Transportation and Logistics: Beyond the Self-Driving Headlines

    Self-driving cars get the headlines, but most of the AI changing transport and logistics today is far less dramatic — and far more useful right now. It decides which van takes which parcel, predicts when a lorry needs servicing, estimates arrival times, and matches drivers to jobs. For small fleets, couriers, taxi and private hire firms, and businesses that deliver their own goods, these tools are increasingly affordable.

    Route optimisation

    Planning routes for several vehicles with many stops, time windows, vehicle capacities, and driver hours is a complex puzzle. Route optimisation software uses algorithms and live traffic data to find efficient routes and re-plan when things change. Typical benefits include fewer miles, lower fuel costs, more stops per day, and more reliable arrival times.

    Small delivery businesses often start with an app that takes a list of addresses and produces an optimised route for each driver, updating as traffic changes.

    Dispatch and job allocation

    For taxi, private hire, and courier operators, dispatch systems increasingly use AI to allocate jobs to the most suitable driver — considering location, traffic, vehicle type, and fairness across drivers. Demand forecasting predicts busy periods, such as airport arrivals, events, weather changes, or Friday nights, so operators can have enough drivers available. Accurate ETAs and automatic customer updates cut down “Where’s my driver?” calls.

    Fleet telematics and predictive maintenance

    Telematics devices record location, speed, fuel use, engine data, harsh braking, and idling. AI analyses this data to:

    • predict mechanical issues before breakdowns;
    • identify fuel waste;
    • coach drivers on safer, smoother driving;
    • support insurance discounts in some cases.

    Dashcams with AI can detect distraction or fatigue and provide evidence after incidents. Drivers should be told clearly what’s monitored and why, and data protection rules apply.

    Warehouses

    Inside warehouses, AI supports demand forecasting, stock placement (putting fast-moving items nearer packing areas), picking routes, and robotics. Large operators use autonomous mobile robots and automated storage systems. Smaller warehouses can benefit from warehouse management software with forecasting and smarter slotting, without major robotics investment.

    Last-mile delivery

    The final leg to the customer is often the most expensive part. AI helps with delivery time slot planning, predicting failed deliveries, choosing between parcel lockers and home delivery, and giving customers accurate delivery windows. Delivery robots and drones are being trialled in some areas, but remain limited for most businesses.

    Freight and supply chains

    Freight platforms use AI to match loads with available lorries, reducing empty return journeys. Shipping and logistics firms use it to predict delays at ports and borders. For small importers and exporters, better tracking and delay predictions help set realistic expectations with customers.

    What about self-driving vehicles?

    Autonomous vehicles are advancing, with robotaxi services operating in some cities abroad. In the UK, the Automated Vehicles Act 2024 created a legal framework for self-driving vehicles, with rules being introduced in stages. For most small transport businesses, autonomous vehicles are not an immediate planning factor, but driver-assistance technology in new vehicles is already improving safety.

    Risks and considerations

    • Driver relations: constant monitoring can feel intrusive. Be transparent and use data fairly.
    • Data quality: poor address data leads to poor routes.
    • Over-optimisation: squeezing too many stops into routes can push drivers into unsafe behaviour.
    • Cyber security: connected fleets and dispatch systems need protection.
    • Algorithmic fairness: job allocation should be explainable to drivers.

    Where a small fleet or transport business can start

    1. Use route optimisation for multi-drop work.
    2. Add telematics to track fuel, idling, and maintenance needs.
    3. Automate customer updates with live ETAs and delivery notifications.
    4. Review booking data to spot peak times and plan driver availability.
    5. Measure before and after — miles, fuel, jobs per shift, and complaints.

    Frequently asked questions

    Is route optimisation worth it for a few vans?

    Often yes. Even small fleets can save meaningful time and fuel, and many apps charge per vehicle per month.

    Can AI help taxi companies?

    Yes: dispatch, demand forecasting, pricing, ETAs, and customer messaging all benefit. See our briefing on AI customer service.

    Are self-driving taxis coming to the UK?

    The legal framework is in place and pilots are planned, but widespread services will take time and depend on safety approvals.

    Related reading