Generative engine optimisation (GEO) means shaping your website and wider online presence so AI answer engines, such as ChatGPT search, Google's AI Overviews and AI Mode, Microsoft Copilot and Perplexity, can find, trust and cite you. Most small businesses don't need a separate GEO programme. They need sound SEO plus a few checks: crawler access, clear facts, and mentions beyond their own site.
The term comes from a 2023 research paper, "GEO: Generative Engine Optimization", later presented at the KDD 2024 conference. The authors built a benchmark of queries and tested ways of rewriting web content; they reported that some methods, such as adding citations, quotations and statistics, could boost a source's visibility in generative engine responses by up to 40%, with results varying a lot by subject. That's a lab result on a test engine, not a promise about ChatGPT or Google. It's a useful idea, and it has since become a label that agencies sell under.
What the research tested, and what it didn't
The paper's contribution was to measure something new: when an AI engine answers a question using several web sources, how prominently does each source appear in the answer? The researchers then rewrote sources in different ways and measured the change. Methods that made content more quotable and better evidenced tended to help. Methods resembling old-fashioned keyword stuffing tended not to.
What it didn't test: your local barber shop in a real commercial assistant, where the answer also depends on reviews, maps data, directories and which pages the engine's search step retrieves in the first place. Treat the finding as confirmation of common sense (clear, specific, well-supported content gets used), not as a formula.
How AI answer engines decide what to cite
Most AI answer engines work in two steps, and understanding them removes most of the mystery.
- Retrieval. The engine runs searches behind the scenes to find relevant pages. Google describes AI Overviews and AI Mode using a "query fan-out" technique, issuing multiple related searches across subtopics and sources. ChatGPT search draws on pages crawled by OpenAI's OAI-SearchBot. Microsoft's Copilot generates what it calls grounding queries to fetch web content. If your page isn't in the index these steps search, it can't be cited.
- Synthesis. The model reads what was retrieved and writes an answer, citing some sources. Pages that state the answer plainly, with specific facts and evidence, are easier to use. For recommendations ("which dry cleaner handles wedding dresses?"), the model also leans on what several sources agree about: reviews, directories, local press, your own site.
Two consequences follow. First, most of GEO is ordinary SEO, because retrieval uses search indexes; Google says there are no additional requirements or special optimisations for its AI features beyond normal SEO. Second, the part that is different is the synthesis step: whether your content is quotable, and whether other sources say the same good things about you.
GEO and SEO: what overlaps and what's actually different
| Area | Traditional SEO | What GEO adds |
|---|---|---|
| Crawling and indexing | Googlebot and Bingbot can reach and index your pages | Also check AI search crawlers such as OAI-SearchBot aren't blocked |
| Content | Pages match what people search for | Answers stated in the first lines, specific facts, figures and examples that can be quoted accurately |
| Authority | Links from other sites | Consistent mentions and descriptions across reviews, directories and press, which models treat as corroboration |
| Local | Business Profile, reviews, citations | The same, plus review wording that describes what you're good at |
| Measurement | Rankings, clicks, Search Console | Citations and mentions in AI answers, tracked with a fixed prompt log and Bing's AI Performance report |
| What doesn't help | Keyword stuffing, thin pages | The same, plus llms.txt files and "AI markup" for Google's AI features, which Google says it doesn't use |
Does your business need it? A decision by business type
How much GEO matters depends on how your customers choose. The more they research and compare before buying, the more AI answers shape the shortlist.
| Business type | How customers choose | GEO priority | What to focus on |
|---|---|---|---|
| Walk-in local service (barber, dry cleaner, nail salon) | Nearby, open now, good reviews | Low to medium | Business Profile, reviews, accurate hours. The five checks below take an afternoon. |
| Considered local purchase (optician frames, garden design, pet grooming for anxious dogs) | Research, compare, then visit | Medium | Answer pages for pre-purchase questions; reviews that describe specifics |
| Online shop | Discovery and comparison, increasingly inside assistants | Medium to high | Product data, clear policies, reviews; see the shopping-specific guidance |
| Specialist or business-to-business service | Long research, "who are the options for…" questions | High | Expert pages, case examples, mentions in trade press and directories |
For a walk-in business, "GEO" mostly means doing the local basics well; how AI Overviews are changing local search covers that side in detail. For an online shop or a specialist service, it deserves a regular slot in your marketing routine.
Five checks that cover most of GEO for a small business
1. Let the search crawlers in
Check your robots.txt file (yoursite.com/robots.txt). Some website builders and security plugins block AI crawlers by default. OpenAI's documentation separates its crawlers: OAI-SearchBot surfaces sites in ChatGPT search answers, and sites that block it won't be shown there; GPTBot collects content that may be used to train models; ChatGPT-User visits pages when a user asks ChatGPT to, and robots.txt rules may not apply to it. Google-Extended controls whether Google may use your content to train future Gemini models, and Google says it doesn't affect inclusion or ranking in Google Search. A common choice for a business that wants to be found but not used for training:
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /
User-agent: Google-Extended
Disallow: /Whether to block training crawlers is your call; allowing them isn't required to appear in search-style answers. Check the result with a robots.txt tester after any change.
2. Publish a clear facts page
AI answers go wrong when facts are scattered or contradictory. One page (often your About or Visit page) should state plainly: what you do, services and price ranges, hours, where you are and who you serve, how to book, and anything that sets you apart, with a concrete example. Keep it identical to your Business Profile.
A filled-in example for an illustrative nail salon's Visit page, written as short labelled lines so both people and AI tools can pick out each fact:
What we do: gel and BIAB manicures, nail art, pedicures, nail repair after acrylics. Prices: gel manicure from $38, BIAB from $45, pedicure from $42; full list on our prices page. Hours: Tuesday to Saturday, 9am to 6pm; late until 8pm Thursday and Friday; closed Sunday and Monday. Booking: online at any time, or by phone during opening hours; walk-ins when there's space. Access: step-free entrance, two parking spaces at the back. What's different: we assess nail health before every first appointment and will recommend a repair course instead of new extensions if your nails need it.
Nothing clever, and that's the point. When an assistant is asked "which salon near the station is open late on Thursday and can fix damaged nails?", every fact it needs is on one page.
3. Write answer-first pages for real questions
List the questions customers ask before buying and give each a page or a section that answers in its first two sentences. Microsoft's guidance for its AI Performance report points the same way: clear headings, tables and FAQ sections help AI systems reference content accurately, and examples, data and cited sources build trust when content is reused. Add a date and keep pages current.
4. Get mentioned beyond your own site
Models are more confident recommending a business that several independent sources describe consistently. For a small business, realistic sources are: reviews on the platforms your customers use, trade and local directories, suppliers' "where to buy" pages, local press or community sites, and industry associations. You can't manufacture these, but you can ask for them: a garden design firm can ask the nursery it buys from to list it as a recommended designer.
5. Keep it fresh and consistent
Out-of-date prices and hours travel into AI answers. Review your facts page, profile and top pages quarterly. Bing supports IndexNow, a way of notifying search engines when pages change, which Microsoft recommends so AI systems use the latest version; many website platforms and plugins offer it.
A barber shop tests ten prompts across three assistants
An illustration of the simplest useful GEO measurement. A three-chair barber shop wrote ten prompts its customers might plausibly use, ran each once in ChatGPT, Google (noting any AI Overview or AI Mode answer) and Copilot, and logged whether the shop was mentioned and what sources were cited. It took about 50 minutes.
| Prompt | ChatGPT | Google AI | Copilot | Sources cited |
|---|---|---|---|---|
| barber for skin fades in [town] | Mentioned | Map pack, not AI | Mentioned | Review sites, the shop's site |
| barber good with autistic children [town] | Not mentioned | Not mentioned | Not mentioned | A competitor's page, a parents' forum |
| walk-in barber open Sunday [town] | Mentioned, wrong hours | Mentioned | Not mentioned | An old directory listing |
| how much is a beard trim at [shop name] | "Price not available" | "Price not available" | Guessed a figure | None |
| (six more prompts) | Mentioned in 4 of 18 answers | |||
The fixes were specific and cheap:
- Quiet appointments. The shop already offered quiet, first-thing appointments for children who find noise difficult, but said so nowhere. It added a short page explaining how they work, and a line on its profile.
- Wrong hours. The old directory listing still showed pre-2024 Sunday hours. Corrected.
- Prices. A price list page went up, with the same prices on the profile. Copilot's guessed figure had been $5 too low, which would have caused awkward conversations at the till.
Two months later, the rerun showed the shop mentioned in 11 of the 30 answers instead of 7 of 30, including the quiet appointments question in two of the three assistants. Illustrative numbers, but a realistic pattern: the biggest gains come from publishing facts you already had but never wrote down.
Rewriting a page so an AI can quote it
The research idea in practice. A dry cleaner's wedding dress page, before:
Your wedding dress holds precious memories, and at our family business we treat every gown with the love and care it deserves. Using specialist techniques perfected over decades, we'll restore your dress to its former glory, ready to be treasured for years to come.
After:
We clean wedding dresses in about three weeks, from $150 for a simple gown to $280 for heavily beaded or long-train dresses. Every dress is inspected before cleaning, with loose beads secured by hand. Stains such as make-up, grass and red wine usually come out; old yellowing sometimes can't be fully reversed, and we'll tell you before we start. Dresses come back in an acid-free box if you want to store them.
The second version answers "how long", "how much", "what can you fix" and "what can't you", each in a sentence a model can quote without distorting it. It also admits a limitation, which reads as more trustworthy to people and gives an AI a reason to recommend the shop for honest advice. None of it is written "for AI"; it's what a customer wanted to know. (The prices are illustrative; publish your own.)
Measuring AI visibility without expensive tools
- A fixed prompt log. The same 10 to 20 prompts, monthly, across two or three assistants. AI answers vary between runs, so look at the trend over months rather than any single result.
- Bing Webmaster Tools' AI Performance report. Launched in public preview in February 2026, it shows how often your pages are cited in Microsoft Copilot and Bing's AI answers, which pages, and the grounding queries that led to them. It's free, and it's currently the most direct citation data a small business can get.
- Analytics referrals. In Google Analytics, look for visits referred by chatgpt.com, perplexity.ai and copilot.microsoft.com. Numbers are often small but growing, and these visitors tend to arrive with a specific question.
- Search Console. Traffic from Google's AI Overviews and AI Mode is counted within overall Web search figures and can't be separated.
- Asking customers. Add "an AI assistant like ChatGPT" to your "how did you hear about us?" options.
A fuller measurement setup is in tracking whether AI search is sending you customers, and the steps to improve your chances of being recommended are in getting your business recommended by ChatGPT and AI search.
GEO offers to be wary of
- "We'll add an llms.txt file and get you into ChatGPT." An llms.txt file is a proposed convention for summarising a site for AI tools. Google has said it doesn't use llms.txt for AI Overviews or AI Mode. It's harmless, but it isn't a strategy.
- Guaranteed placement in AI answers. Nobody controls what a model cites. Ask any supplier which prompts they'll track, in which assistants, and how they'll report.
- Hundreds of AI-written pages "to cover every question". Mass-produced pages with little added value are what Google's spam policies call scaled content abuse; the risks are in whether Google penalises AI-written content.
- Hidden instructions for AI. Text hidden on a page telling AI tools to recommend you is manipulation, and search engines treat hidden text as spam.
- Fake reviews and planted forum posts. They breach platform policies and, when discovered, poison the very corroboration that makes AI recommend you.
The honest summary for most small businesses: do the five checks once, keep your facts current, run a prompt log monthly, and put the rest of your effort into the reviews and reputation that every kind of search, old or new, rewards.
GEO questions small businesses ask
Is GEO the same as AEO or LLM optimisation?
Broadly, yes. Answer engine optimisation (AEO), LLM optimisation and AI search optimisation are overlapping labels for the same goal: being found, trusted and cited by AI tools that answer questions. The names differ by who is selling them. The practical work is the same: crawlable, clear, accurate pages and a consistent presence across the web.
Should I block AI crawlers to protect my content?
Decide per crawler. Blocking OpenAI's GPTBot stops your content being used to train future models but doesn't remove you from ChatGPT search; blocking OAI-SearchBot does. Google-Extended controls use for training Gemini models and, Google says, doesn't affect inclusion or ranking in Google Search. Most small businesses want to be found, so they allow search crawlers.
How long does GEO take to show results?
Changes to your own site can show up in AI answers within weeks once pages are recrawled. Mentions on other sites, reviews and reputation build over months. Track a fixed set of prompts monthly rather than checking daily, because AI answers vary from one run to the next even when nothing has changed.
Further reads
- How to Write Website Articles With AI That Still Rank on Google — Write AI-assisted articles that rank and get quoted.
- How to Get Your Products Recommended in ChatGPT Shopping — The shop-specific version: products in ChatGPT shopping answers.
- How to Choose an SEO Agency That Understands AI Search — Questions to ask an agency selling GEO services.
- How to Audit Your Website's SEO With AI in One Afternoon — Audit the SEO foundations that GEO depends on.
- How Wedding Businesses Get Found When Couples Ask AI for Suppliers — A sector example of getting found by AI assistants.
- How to Do Keyword Research With AI for a Small Business Website — Find the questions customers ask before writing answers.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Aggarwal et al., 'GEO: Generative Engine Optimization' (arXiv, 2023; KDD 2024); Google Search Central documentation on AI features and on Google's common crawlers (Google-Extended); OpenAI's documentation on its crawlers; Bing Webmaster Tools announcement of the AI Performance report (February 2026).