How to get cited by AI: a practical starting point
Make useful business information accessible to AI search, separate search crawlers from training controls, and measure citations without promises.
On this page
You can improve whether an AI search service can find and understand your website. You cannot make it recommend your business, control which sources it cites or guarantee that a citation will produce an enquiry.
For a small business, start with an accurate, accessible website that answers real buying questions. Then check the relevant search controls and collect a repeatable baseline. Buying a special "AI optimisation" file before doing that is the wrong order.
Separate eligibility from selection
Google says its ordinary SEO requirements apply to AI Overviews and AI Mode. A supporting page must be indexed and eligible for a search snippet; meeting those conditions does not guarantee inclusion. It requires no special AI markup or text file. See Google's guidance for AI features.
That is useful documentation for Google's products. It is not a description of every assistant's source-selection system. Treat a claim that one page format "makes all AI engines cite you" as a claim requiring evidence.
A practical first check is to inspect your important service or information page in Search Console. Confirm that the public version is accessible, the intended canonical URL is selected and an accidental noindex instruction is not excluding it.
Search access and model training are different decisions
Do not copy an "allow every AI bot" configuration without understanding it.
| Control | What it concerns |
|---|---|
| OAI-SearchBot | OpenAI's crawler for ChatGPT search results. |
| GPTBot | Crawling for possible training of OpenAI's foundation models. This permission is separate from search. |
| ChatGPT-User | Certain visits initiated by a ChatGPT user; it is not the automatic search-indexing crawler. |
| Googlebot | Crawling for Google Search, including its AI search features. |
| Google-Extended | Certain Gemini training and grounding uses; it does not control inclusion or ranking in Google Search. |
These distinctions come from OpenAI's crawler documentation and Google's crawler documentation. Check other providers' current documentation separately.
Also check hosting and firewall rules: a permissive robots.txt file does not help a crawler that receives an access-denied response. Verify crawler identity before changing security rules. Make deliberate access choices; do not weaken site security to chase mentions.
Give a prospective customer something worth finding
Our editorial recommendation is to start with a useful buying decision, rather than a list of prompts you hope to rank for.
Consider a fictional Ipswich repair business. "We deliver excellent service" tells a customer little. A page explaining the equipment it repairs, its genuine service area, how a fault is assessed, what affects the price and how to request a booking helps that customer decide whether to make contact.
The same approach applies to more technical products:
- Explain what the product does and who it suits.
- State limitations and requirements alongside benefits.
- Give current specifications, supported integrations or a worked example.
- Identify who produced the information and when important details were checked.
- Support factual claims with original evidence or a directly relevant source.
These are recommendations for useful publishing, not a tested formula for winning citations. Keep related questions together when they belong to the same decision. A separate, near-identical page for every wording adds little.
For case studies, distinguish measured results from the owner's account. A screenshot of search clicks does not establish sales revenue or prove which change caused growth.
Structure helps readers; it does not buy endorsement
Use clear headings, a direct opening answer and descriptive internal links. Keep important specifications available as text, with helpful images alongside them. A comparison table is useful when the comparison itself is fair.
If you publish structured data, make it match the visible page. Never invent reviews, awards or business locations for markup. Google's AI guidance does not prescribe a special citation schema.
An llms.txt file can be an optional navigation aid for systems that choose to use it. Its presence is not proof that a provider has read your site, and it is not a substitute for ordinary access and indexing.
Measure a sample, and label it as a sample
Choose a small set of realistic questions from actual customer conversations. Record the product used, date, wording, relevant location, named businesses and linked source URLs. Repeat the same set periodically. An answer can change across runs, products and users, so one favourable screenshot is not a market-wide visibility score.
Keep three outcomes separate:
- Mention: the business or product is named.
- Citation: the answer links to a source page.
- Business result: someone visits, makes a suitable enquiry or buys.
Google includes its AI search-feature traffic in Search Console's Web reporting; do not relabel all search growth as AI growth. Referral data and an optional "How did you find us?" question can add context, but attribution will remain incomplete.
Start with one useful page, correct its access problems, document the baseline and improve the answer. Read how search works next, or explore the AI search reading path.
Need implementation help? Our sister agency Suffolk Web offers AI visibility work. The commercial relationship is explained on our about page.
