Why you rank first on Google and AI still recommends your competitor instead

This is for founders and marketing leads who have done the SEO work, hold position one for the terms that matter, and have just watched an AI assistant name three competitors and not them.

What is the difference between ranking and being cited?

Ranking is a position in a list of results. Citation is being named inside an answer, either as a recommendation or as a source the model leans on to build one.

A search engine returns options and lets the user choose. An assistant makes that choice on the user's behalf and hands back a shortlist, usually three or four names. There is no page two. If you are not on the shortlist you have not lost by a place, you are absent from the conversation entirely.

And that shortlist is built from a far wider pool than your website.

Why does an AI assistant skip a site that ranks first on Google?

Usually because the model has nothing to go on except your own marketing copy, and it discounts that heavily.

When an assistant answers a question like "who are the best corporate tax advisers in Bristol", it is looking for corroboration. Independent sources that name you. Roundups, directories, reviews, trade press, a forum thread where someone recommended you without being asked. Those carry weight precisely because you did not write them.

Your homepage says you are the trusted specialist in your field. So does every competitor's homepage. The model cannot break that tie from your site alone, so it reaches for the sources that already broke it.

There is a second failure sitting underneath the first. Some sites never enter the pool at all.

Retrieval is whether your content can be found and read at the moment an answer is being assembled. Citation is whether it then gets chosen. A site can fail at retrieval and keep ranking perfectly well on Google, because nothing in your analytics tells you it happened.

What keeps showing up in these AI visibility audits?

The same pattern, most weeks.

Across the free Visibility Snapshots I have run this year, mostly for finance and professional services firms, the businesses that are invisible in AI answers are rarely the ones with bad websites. They usually have solid sites, genuine rankings and years of Google-focused work behind them. The technical foundations are fine.

What they do not have is a footprint anywhere else. Nothing in the sector roundups their buyers actually read. No named partner or founder with a traceable public record. Nothing on LinkedIn or Reddit that a model could pick up as third-party evidence that this firm exists and is any good.

Meanwhile the competitor getting named in every answer is often ranking below them, and occasionally has the worse website. It got into four listicles.

That is the whole difference. Not budget, not domain authority. Corroboration.

Does schema or an llms.txt file fix AI search visibility?

They help, and they are oversold.

Schema is structured code that tells a machine what your page covers, who wrote it and what your business does. An llms.txt file is a plain text file at the root of your site that gives AI crawlers a guide to your content. Both make you easier to read accurately, and both sit inside what the industry now calls Answer Engine Optimisation, or AEO.

Neither one manufactures authority. If a model has already decided you are not part of the answer, tidier markup will not change that decision. Where they earn their place is once you are a candidate, because a machine that can parse you correctly is far more likely to describe you correctly.

Get them right. Do not expect them to be the fix.

What should I actually do about it?

Four things, in this order, and the order is the point.

1. Ask the assistants the questions your buyers ask. Write down who gets named and, more importantly, which sources get cited underneath the answer.
2. Look at where those cited sources live. Roundups, directories, review sites, threads, trade publications. That list is now your target list.
3. Get into three of them. Pitch the roundup, submit to the directory, answer the question in the community properly. This is slow, unglamorous work and it is the work.
4. Then fix the technical side so a crawler can read you cleanly when it arrives.

Most businesses do step four and skip one to three, because step four can be bought and finished in a fortnight. The other three cannot.

The part that has not changed

An assistant recommending a supplier is running a compressed version of what a person does when they ask three people they trust before choosing. It is weighing who else vouches for you.

That was how buying decisions worked before any of this. The mechanism now sits inside a model instead of a conversation, and it moves faster, but what it is looking for is the same thing it was always looking for. Evidence that someone other than you thinks you are good.

Optimise for the machine all you like. It is still asking a human question.