You rank on page one for your main term and an assistant still recommends four competitors and not you. This is the most common complaint we hear about AI search, and the reason is structural: your ranking is not an input the assistant reads.
How an assistant actually chooses
Roughly, it does three things. It works out what kind of answer the question needs. It gathers candidate sources, sometimes from a live search, sometimes from what it already learned during training. Then it writes an answer, preferring claims that are specific, current, and repeated across more than one source.
Nothing in that sequence rewards being the top blue link. It rewards being the clearest and most corroborated statement of a fact.
The four things that do carry over
-
Being nameable at all
You need an unambiguous entity: one consistent company name, address and phone number, an Organization node in your structured data on every page, and profiles elsewhere that agree. If an assistant cannot resolve who you are with confidence, it will not name you.
-
Saying something specific enough to quote
Ranges, timeframes, named integrations, stated process. Hedged marketing copy gives an assistant nothing to lift. This is where most sites fail, and it is entirely within your control.
-
Corroboration off your own site
Assistants weight agreement between independent sources. Directory listings, review platforms, trade bodies and press mentions are what turn your claim into a fact. A site with no third-party footprint is a single unverified source.
-
Being current
Assistants prefer recent information and many now check live. A page last updated in 2023 competes badly against one updated last month making the same claim.
The uncomfortable part
Two of those four are off-site. You can write the perfect page and still not be named, because nothing else on the internet confirms you exist. For a new business this is the actual bottleneck, and no amount of on-page work substitutes for it.
That is why this work is closer to PR and citation-building than to traditional SEO, and why anyone selling you AI visibility as a purely on-page exercise is selling half the job.
What to do, in order
One. Fix the entity. One canonical name and address everywhere, Organization schema on every page rather than just the homepage, and a llms.txt that states plainly what you do.
Two. Rewrite your key pages to answer the buying question in the first paragraph, with specifics rather than adjectives.
Three. Get listed and reviewed where your sector is discussed, so your claims are corroborated by somebody other than you.
Four. Measure it. Track the questions your buyers actually ask, across several assistants, over time. You cannot see this in analytics because there is often no click.
See AI search visibility, or the benchmark of who assistants actually name.
Questions we get asked
How long does it take to start being cited?
Weeks to months, not days. Models need to re-crawl or re-index, and corroboration takes time to accumulate. Anyone promising results next week does not understand the mechanism.
Does paying for ads help?
No. Assistant answers are drawn from content, not ad inventory. Ads may drive traffic but they do not make you quotable.
Why does it name a competitor with a worse website?
Almost always corroboration. They are listed, reviewed and mentioned in more places, so the assistant is more confident they exist and can be recommended safely.
Can we ask an assistant to include us?
No, and anyone offering to place you in ChatGPT is describing something that does not exist. You influence it by being the clearest, best-corroborated source, and by nothing else.