In short
- Nobody can guarantee that an AI assistant will mention your business. Anyone selling a guaranteed citation is selling something they cannot deliver, and that is the honest headline answer.
- What you can do is make yourself easy to cite: let AI crawlers reach your pages, state verifiable facts in plain language, mark them up with structured data, and keep your name, address and phone number identical everywhere.
- Assistants lean on what other sites say about you. Directories, trade bodies, local press and review platforms corroborate your claims, and corroborated claims get repeated.
- llms.txt is a proposed convention, not a ranking factor. It costs almost nothing to publish, is not honoured by every provider, and should sit at the bottom of your list rather than the top.
What is actually happening when an assistant answers
Two mechanisms are at work. Some answers come from what the model absorbed in training, a slow closed process you cannot influence. Most answers that matter commercially come from retrieval: the assistant searches, fetches a handful of pages, reads them and writes an answer with links. Google’s AI Overviews work broadly this way, drawing on the index behind ordinary search results.
That second mechanism is the one worth working on, and the implication is reassuring. If your pages are findable, accurate and easy to read, you are most of the way there. The work is less exotic than “generative engine optimisation” suggests: it is largely good search work with the emphasis shifted from ranking to being quotable.
Let the crawlers in
An assistant cannot cite a page it cannot fetch. This is the commonest reason a perfectly good business is invisible in AI answers, and it is usually accidental.
- Check your robots.txt. A blanket block from years ago, or a staging rule that survived launch, will exclude AI crawlers along with everything else.
- Decide deliberately about the named AI user agents. OpenAI publishes its crawler names and Google documents Google-Extended. Blocking them means opting out of citation.
- Make sure content exists in the HTML. If key text only appears after JavaScript runs, some fetchers see an empty page.
- Keep the site fast and stable. Fetchers time out, and a page that fails to load never gets quoted.
- Check your firewall or security plugin. Aggressive bot protection often blocks these crawlers silently.
That check takes minutes and belongs in any competent technical audit. Our web development work covers it as standard, because the same faults hurt ordinary search too.
Write things that can be quoted
Language models extract claims. A page of clear, self-contained facts gives them something to extract. A page of atmospheric marketing copy does not.
In practice that means answering the question in the first sentence of a section rather than the last, writing “we cover Greater Manchester and work with clients across the UK” instead of “wherever you are, we are with you on the journey”, and putting prices, service areas, hours and qualifications in unambiguous sentences rather than implying them. Write as a primary source: you are the authority on your own business, so state facts directly.
Dates matter too. A page that says when it was last reviewed, and genuinely is, beats one of uncertain vintage. So does structure: descriptive headings, short paragraphs and a real question-and-answer section make a passage easy to lift without distorting it.
Structured data and consistent business information
Structured data does not force an assistant to mention you, but it removes ambiguity about who you are. Organisation and LocalBusiness markup ties your name, address, phone number, hours and profiles together in machine-readable form, and FAQ markup does the same for questions. Google documents all of it publicly, and it serves conventional search features too, so the effort is never wasted.
Consistency matters more than people expect. If your phone number appears three ways across your site, your Google Business Profile and an old directory listing, a system summarising your business must pick one, may pick wrongly, or may hedge and name a competitor whose details agree with themselves. Pick one canonical version and enforce it everywhere. Dull work, and the part that most reliably improves how accurately you are described.
Third-party corroboration
Assistants weigh independent sources. Your own site establishes what you claim; everything else establishes whether anyone agrees.
- A complete, actively maintained Google Business Profile, with categories that match what you actually do.
- Reputable directories and trade bodies in your sector, with details matching your site exactly.
- Genuine reviews on platforms people trust, answered rather than left to sit.
- Public registrations, such as a Companies House record, giving a verifiable anchor for the business identity.
- Mentions and contributions on sites that are not yours, earned rather than bought.
None of this is new. It is the corroboration that has underpinned local search for years, which is why businesses already doing it well tend to appear in AI answers without doing anything special. Steady content marketing is how most small businesses build that footprint.
Where llms.txt actually fits
llms.txt is a proposed file at the root of a domain offering a curated, plain-text map of a site for language models. The idea is sensible: point a machine at your most useful pages rather than making it wade through navigation and scripts.
Be clear about its status. It is a community convention, not a standard adopted by search engines, and it is not a ranking factor. Providers have not all committed to honouring it, and publishing one will not by itself change whether you are cited. It is cheap and harmless, so publish one if you like, but after the crawlability, clarity, markup and consistency work rather than instead of it. Anyone presenting llms.txt as the key to AI visibility has the priorities backwards.
Measuring something this hard to measure
Assistant answers are personalised, non-deterministic and often unlinked, so rank tracking does not apply. Two people asking the same thing can get different answers from different sources. What you can do is watch referral traffic from assistant domains, monitor branded search volume, ask new enquirers how they found you, and periodically put your customers’ questions to a few assistants and record who gets named.
Treat that as a directional signal, not a metric. It tells you whether you are in the conversation, not your position in it. Proper analytics and tracking at least ensures that when an assistant does send someone, you know about it.
Frequently asked questions
Can you guarantee my business will be cited by ChatGPT?
No, and neither can anyone else. Citation depends on the model, the wording of the question, the retrieval system and what else exists on the web at that moment. You can improve the odds substantially. You cannot buy the outcome.
Is AI search optimisation different from normal SEO?
Mostly it is the same work with a different emphasis. Crawlability, accurate content, structured data and corroboration serve both. The real difference is writing for extraction rather than for a click, and caring about being described correctly as much as being ranked.
Should I block AI crawlers to protect my content?
It is a legitimate choice and some publishers make it, but understand the trade-off: a blocked site cannot be cited, so you are choosing protection over visibility. For most small businesses trying to be found, allowing them is more sensible.
Do I need a separate AI content strategy?
No. You need your existing content to be clearer, better evidenced and easier to verify. A separate strategy usually produces thin pages written for machines, which serves neither the machines nor the people who read them.
Key takeaways
- No agency can guarantee AI citations. Be sceptical of anyone who says otherwise.
- Crawler access comes first. Nothing else matters if fetchers cannot reach your pages.
- Plain, factual, well-structured writing is what gets extracted and quoted.
- Structured data and identical business details everywhere remove the ambiguity that gets you skipped.
- Independent corroboration carries more weight than anything you say about yourself.
- llms.txt is emerging, optional and unproven. Do the fundamentals first.
If you would like someone to check whether AI crawlers can reach your site, whether your structured data is right and whether your business details agree across the web, that is ordinary work for us rather than a special project. Packages start at £399 a month with fixed published pricing, no long-term contract and a reply within 24 hours. We are based in Oldham, registered at Companies House, and work with businesses across Greater Manchester and the UK. Our AI solutions page explains how we approach it.