AI Can't Recommend What It Can't Understand

An assistant recommends what it can describe with confidence. If what you do is fuzzy or buried, the model hedges or skips you, and clarity about what you offer is quietly becoming a ranking factor.

A recommendation is a claim the model is willing to make

When you ask ChatGPT, Perplexity, or Gemini for the best tool for a job, you are not asking it to list pages that match your words. You are asking it to vouch for something. A recommendation is a claim the model makes on your behalf, in front of a user who will act on it.

Models are cautious about claims they cannot support. If the picture it holds of you is thin or contradictory, naming you is a risk: it might be wrong, and being wrong in a confident sentence is the failure these systems are tuned to avoid. So it does the safe thing. It names the options it can characterize cleanly, and leaves you out.

How the model builds its picture of you

The picture is assembled from text: your own pages, third-party descriptions, directory listings, reviews, whatever it has read that mentions you. From that it forms a compressed summary, roughly “this is a company that does X, for Y, and differs by Z.”

Two things degrade that summary. The first is vagueness: a homepage that says you “help businesses grow” gives the model nothing it can match to a specific request. The second is inconsistency: when your homepage, your about page, and a listing describe you three different ways, the model cannot settle on one confident line, so it holds all three loosely.

Where a description is specific and corroborated across sources, the model’s confidence is high. Where it is generic or self-contradictory, confidence drops, and low confidence reads, at answer time, as silence.

Retrieval finds you; description decides you

Grant the fair objection: surely the model surfaces the best match regardless of how you phrase things. Retrieval does find candidates, and a page that matches the query can enter the running on relevance alone.

Retrieval and recommendation are different steps, though. Getting pulled into the candidate set is not the same as getting named in the answer. Among the candidates, the model prefers the ones it can describe in a sentence without hedging, because the answer it writes has to stand up on its own. Clarity is the tiebreaker, and the tie is increasingly where the whole contest is decided.

What clarity looks like in practice

Clarity here is not a tone. It is a set of facts a machine can lift and repeat without qualification.

One description, everywhere the same. A single plain line of what you do, and for whom, appearing consistently across your homepage, your about page, and any listing that carries you. The model rewards agreement between sources, so give it agreement to find.

Specific enough to match a request. “We do X for Y” beats “we help businesses grow.” The test is whether a stranger could read the line and know which requests you belong in the answer to.

Corroborated, not only asserted. A claim the model sees on your own site alone is weaker than the same claim echoed by a review, a directory, or a mention elsewhere. Self-description is evidence; independent agreement is better evidence.

Scoped so a passage stands on its own. State what you do in a self-contained sentence near the top, not spread thin across a paragraph of warm-up. The model quotes passages, not whole pages.

None of this is writing for the machine at the reader’s expense. A description crisp enough for a model to repeat is a description clear enough for a hurried human to act on. The two audiences want the same thing.

Clarity is the ranking factor now

For years the visibility question was where you ranked. The newer question is whether an assistant can say what you do in one confident sentence. If it can, you get named. If it cannot, you get a hedge or nothing, however good the underlying offer.

That makes clarity a ranking factor, not a branding nicety. It is also the rare one you hold in your own hands.

Run the free scanner at wpvisibility.com to see how clearly your site describes itself. It grades your AI visibility in about a minute. WP Visibility is $99 a year for unlimited sites, with client sites included and a 30-day refund.