A customer who used to search “best plumber in Salford” now sometimes asks ChatGPT “who should I call for a boiler leak in Salford, and roughly what will it cost?” — and acts on the answer without visiting a single website. Google’s AI Overviews do the same thing at the top of ordinary searches. The click you used to compete for is being spent inside the answer, and the question for a small business is no longer only “do I rank?” but “am I the source being quoted?”. That is what AI SEO — or generative engine optimization, if you prefer the longer name — actually means.
The good news, having tested this across client sites: there is very little new work. Almost everything that earns an AI citation is something a careful SEO would have done anyway, done more literally. This is the list, in the order I apply it.
Where AI answers actually come from
Four sources, roughly in order of weight for a local or service business:
- Pages the engine can retrieve and parse quickly — usually pages already ranking well in classic search, because the AI engines lean on the same index or their own crawl of it.
- Third-party confirmation — directories, review platforms, trade bodies, local press, Reddit and forum threads. AI answers love sources that agree with each other.
- Structured facts — schema markup, your Google Business Profile, consistent name/address/phone across the web.
- Clear, quotable passages — a two-sentence answer directly under a heading beats the same fact buried in paragraph nine.
Notice what is missing: keyword density, word count, and clever copy. Answer engines are extracting facts, not admiring prose.
The nine changes that earn citations
-
Answer first, elaborate after.
Under every H2 that asks a question, put the answer in the first 40–60 words, in plain sentences that stand alone if lifted out of the page. Then expand. This single habit does more for AI citation than anything else on this list — and it improves the page for humans skimming on a phone.
-
Write the headings the way people ask.
“How much does a boiler service cost in Salford?” rather than “Pricing”. Conversational queries are what people type into AI tools, and heading text is one of the strongest matching signals available.
-
Be specific and dateable.
Numbers, prices or ranges, timescales, place names, “as of September 2026”. Vague marketing copy is unquotable; a sentence with a figure in it is exactly what an answer engine wants. Review those figures quarterly — stale specifics are worse than none.
-
Add a real FAQ to every important page.
Four to eight genuine customer questions with short, complete answers, marked up as FAQPage schema. This is the most citation-friendly block on a page, and the schema makes the question-answer pairing unambiguous.
-
Get the entity facts consistent.
Organization (or LocalBusiness) schema with your legal name, address, phone, service area, opening hours and social profiles — matching your Google Business Profile and every directory listing exactly. AI engines assemble a picture of your business from these; contradictions make you a risky source to quote.
-
Keep the technical side boring.
Fast pages, server-rendered HTML (content that only appears after JavaScript may not be read), clean headings, no content locked behind interactions, a valid sitemap. The Core Web Vitals work and the speed checklist cover this; retrieval favours pages that load quickly and parse cleanly.
-
Publish the comparison and cost pages you have been avoiding.
“X vs Y”, “how much does X cost”, “is X worth it” — these are exactly the prompts people give AI tools before buying, and most small business sites answer none of them. An honest comparison page that names competitors gets cited more often than a sales page that does not.
-
Earn third-party mentions.
Directories and trade listings with consistent details, a few genuine local links, reviews that mention the service and the town, and — where it fits naturally — being a useful participant in the forums your customers read. Answer engines weight corroboration heavily.
-
Decide what you let crawlers take.
AI crawlers (GPTBot, PerplexityBot, ClaudeBot, Google-Extended and others) can be allowed or blocked in robots.txt. Blocking them protects content from training use but also removes you from citations. For a business that wants customers, allow them; for a publisher selling content, it is a real decision. Do it deliberately, not by copying someone’s robots file.
llms.txt, honestly
An llms.txt file at the root of your site is a plain-text, markdown-style map of your most useful pages with one-line descriptions — robots.txt’s helpful cousin. Support across AI tools is patchy and it is not a ranking factor; anyone selling it as one is overselling. It also costs about an hour, breaks nothing, and gives any tool that does read it a clean summary of what you do and where the real answers live. I add one to client sites for the same reason I add a sitemap.
What a small-business llms.txt contains: the business name and one-line description, then grouped links — services, pricing, service areas, key guides, contact — each with a short description. Keep it under a page, update it when the site structure changes.
How to check whether it is working
| Check | How | How often |
|---|---|---|
| Are you named in AI answers? | Run your ten customer questions in ChatGPT, Perplexity, Gemini and Google AI Mode; record who is named and which page is cited | Monthly |
| Referral traffic from AI tools | Analytics referrers: chatgpt.com, perplexity.ai, copilot, gemini | Monthly |
| Question-shaped queries | Search Console → Performance, filter queries containing “how”, “what”, “best”, “vs”, “cost” | Monthly |
| Impressions up, clicks flat | Search Console trend — the signature of AI Overviews answering for you | Quarterly |
| Facts still true | Prices, hours, service areas on the pages AI quotes | Quarterly |
Set a baseline today, before you change anything. Without it, three months of work is a feeling rather than a result — the same discipline as the rest of your tracking setup.
What to ignore
- “AI SEO” tools that promise rankings inside ChatGPT. There is no ranking to buy; there are sources to be.
- Mass AI-generated content to “feed the models”. Answer engines prefer corroborated, specific sources; a hundred thin pages make you less quotable, not more.
- Prompt-stuffing your pages with hidden text aimed at models. It is the 2005 keyword-stuffing play, and it ends the same way.
- Abandoning classic SEO. Most AI citations come from pages that already rank. The two jobs are one job.
The order I would do this in
- Pick your ten highest-value pages (services, pricing, main guides).
- Rewrite the opening 60 words under each heading to answer directly.
- Add or fix FAQ blocks and FAQPage schema.
- Fix Organization/LocalBusiness schema and make details consistent with the Business Profile and directories.
- Publish one cost page and one comparison page.
- Add llms.txt; decide the AI-crawler policy deliberately.
- Record the baseline, then re-check monthly.
Two afternoons for most small sites. The businesses that do it are the ones being named when a customer asks a machine for a recommendation — which, increasingly, is the first thing that happens in a buying decision.
Further reading: Google: AI features and your website · FAQPage structured data · The llms.txt proposal.
FAQ
What is generative engine optimization (GEO)?
Structuring and writing content so AI answer engines — ChatGPT, Perplexity, Gemini, Google AI Overviews — quote and cite it when they answer a question. It overlaps heavily with good SEO: clear structure, direct answers, schema, and being a source other sites and directories confirm.
Do I need to do anything different from normal SEO?
Mostly you need to do normal SEO more literally. Answer the question in the first 40–60 words under a heading that matches how people ask it, keep facts specific and current, add FAQ and organisation schema, and make sure your business details match everywhere. Those changes lift both classic rankings and AI citations.
Does an llms.txt file help?
It is cheap and harmless: a plain-text map of your most useful pages at /llms.txt, in the spirit of robots.txt. Support is not universal and it is not a ranking factor, but it costs an hour and makes your key pages easy to find for tools that do read it.
How do I know if AI engines are citing my business?
Ask them. Run the ten questions a customer would ask about your service and area in ChatGPT, Perplexity, Gemini and Google AI Mode once a month and record who gets named. Also watch for referral traffic from chatgpt.com and perplexity.ai in analytics, and impressions on question-shaped queries in Search Console.
Want your key pages rewritten to be quotable, the schema fixed and a baseline set so you can see it working? See the SEO and web development services — this is two afternoons of work that decides whether a machine recommends you or a competitor.
