The default sentence

Start with the mechanic. A buyer opens an assistant and asks who they should talk to about a problem you solve. The assistant does not simply read your positioning statement back. Your homepage is one source; the answer may combine it with documentation, third-party coverage, what the model learned in training and other information — and it can contain errors. It produces a sentence or two about you, does the same for two or three others, and hands over a shortlist.

For some buyers, that answer can shape a first impression before they ever reach your site. It has no author, no approval chain and no archive. It is generated per question and per session, and then it is gone.

Why the default sentence is expensive

Because it is not wrong. If it were wrong, someone would notice and correct it. Blandness does not get corrected — it quietly removes you from consideration, and nobody involved experiences it as a decision.

And because it is shared. “A leading provider of AI-powered solutions for the enterprise” describes you and the two companies you are most often confused with. When descriptions are interchangeable, buyers fall back on the tiebreakers available to them: price, familiarity and whoever was mentioned first.

There may also be a compounding effect. Companies described consistently across many sources seem more likely to be described that way again, which would make early clarity worth more than late clarity. That is a reasonable inference from how these systems learn, not something anyone outside the providers can measure precisely.

Two industries, two partial answers

The brand agency’s answer is a better claim: sharper language, a new tagline, a workshop. Sometimes exactly right, rarely sufficient on its own. Assertion is the cheapest signal there is, and readers — human and machine — have reason to discount it.

The AI-search vendor’s answer is a technical fix: structured data, an llms.txt file, FAQ blocks. Worth doing where it is cheap — and not the answer. In a May 2026 sample of 137,210 domains, Ahrefs found about 38,000 with a valid llms.txt; 97% of those files received no requests that month.¹

Both answers are about expression. The harder question is about evidence: what can be checked about you by someone who has never spoken to you. Your own site is a witness statement. Independent sources are the corroboration.

You cannot write the sentence. You can make the accurate version easier to find, check and repeat.

What is likely to move the answer

Three things, not equally weighted, and none of them guaranteed.

Specificity. “Autonomous inspection for offshore wind, with the certifications listed” is hard to confuse with a competitor. “AI-powered industrial intelligence” is a category with a logo attached. Specificity is not a copywriting preference; it is what makes a description distinguishable at all.

Independent sources. One source saying something is an assertion. Several relevant, independent sources saying the same thing is much closer to a fact — for a buyer, and plausibly for a model. We track how many relevant, independent sources state a claim as a planning diagnostic. There is no known threshold that makes a system repeat it, and syndicated copies of one announcement are not independent.

Readability. The page has to be readable by the systems that fetch it. In December 2024 Vercel reported that, in its observations, the major AI crawlers it measured — including GPTBot, ClaudeBot and PerplexityBot — did not execute JavaScript, while Googlebot did.² Some crawlers may therefore miss essential information that appears only after JavaScript runs. The fix is unglamorous: test the HTML they actually receive, alongside access and indexing.

The order matters more than the list

Readability first, because nothing else can be received. Specificity second, because a readable page with nothing distinctive on it just delivers the default sentence more efficiently. Independent sources last and longest, because they take other people’s time and cannot be bought convincingly.

Run the order backwards — chase mentions before deciding the claim — and you spread the default sentence faster. It is a common way money is wasted in this category.

The uncomfortable part

Sometimes a claim does not stick because it is not yet true. Other people will not repeat a claim they cannot check. When we find that, we say it, and the work changes shape: make the claim true, or state it at the strength it has today.

We would rather lose the engagement than amplify something the market has already declined to believe. Not out of virtue — because it does not work.

What this means for a leadership team

Read what is said about you. Not a share-of-voice percentage — the actual answers, for the questions your buyers ask, with the cited sources recorded. In our experience it is often uncomfortable reading.

Decide what should be said instead, and test it: is it true, does it matter to the buyer, and is it supported by evidence an outsider can check? A position does not need to be untrue of every competitor. It needs to be relevant and defensible.

Then build the evidence, starting with what only you can produce: test results stated with their conditions, documented methods, approved customer outcomes, people who will stand behind a claim. Verifiable evidence is the one asset in this market that cannot be counterfeited convincingly.

None of this requires believing that AI will replace search, or any other forecast that may look silly in three years. It requires one observation anyone can check in a few minutes: ask an assistant about your category, and read what it says about you.