AI recommended your brand. Then a follow-up question made it disappear.
There is a comforting story going around: get mentioned once by an AI assistant and you have won a new front in search. New data says the win is thinner than it looks, and the number that describes it had to be corrected upward before anyone saw how thin.
Search Engine Journal reported on July 7, 2026 on new Clovion data finding that 62% of AI brand recommendations vanish after a single buyer question (searchenginejournal.com). Per the outlet, a dropped zero in Clovion's original AI visibility study had hidden numbers the piece describes as more alarming than what was first published, and the corrected figures are what the analysis is built on (searchenginejournal.com). We have not reproduced Clovion's study ourselves, so we are reporting their finding as theirs, with the source attached.
What the finding actually says
Read plainly, it says this: being named by an AI assistant in a first answer is not the same as being the answer that survives the second question.
A shopper asks a broad question. The assistant lists a few brands, and yours is one of them. Good moment. Then the shopper does what shoppers do, they narrow it: "which of these is best for X," "which is cheapest," "which one integrates with Y." Per the reported data, most of the initial recommendations do not make it through that follow-up. The brand that gets named first and the brand that gets recommended after scrutiny are often not the same brand.
That gap is the whole story. A single mention is a snapshot. A purchase decision is a conversation.
The number correcting itself is part of the lesson
We will pause on the dropped zero, because it is on-brand for us in an uncomfortable way. A visibility study published a figure, someone caught a decimal error, and the corrected version was worse than the original. That is not a knock on Clovion for fixing it. Fixing it is the right move. It is a reminder that numbers about AI visibility, including flattering ones, deserve a second look before anyone builds a strategy on them.
We would rather cite a corrected 62% than a clean-looking figure nobody checked. The same skepticism you would apply to a stat in a study is the skepticism worth applying to an AI assistant's confident answer about your brand. Both can be wrong. Both can change on the next run.
What actually holds up under a follow-up
You cannot control what an assistant says on any given day. You can control whether your site gives it something solid to stand on when the questions get sharper. A few things that tend to hold up:
- Specific, checkable claims. "Handles X for Y kind of business" beats "the best solution for everyone." Assistants can quote specifics; vague praise gives them nothing to carry into the second answer.
- Consistent facts across your pages. If your pricing, features, and positioning disagree with each other page to page, the follow-up question is exactly where that contradiction surfaces and costs you.
- Content that answers the narrowing question, not just the opening one. The comparison, the edge case, the "who is this not for." That is the material an assistant reaches for when a shopper drills down.
- Structure a machine can read. If your best answers are locked in images, or only appear after JavaScript runs, the assistant may never see them at all.
None of this is a trick to game an answer engine. It is the same work that makes your site honest and legible to a human who is deciding whether to trust you.
Our take
We build AuditLamp to read your site the way search engines and AI answer engines actually read it, and to tell you the plain truth about what they can and cannot use. Whether your pages give an assistant something specific enough to quote, and consistent enough to defend on the second question, is squarely part of that.
The free scan checks whether your content is structured to be quoted and whether the basic trust signals are there and agree with each other. It will not promise you a permanent seat in an AI answer, because no honest tool can. It will show you, in plain language, whether your site is giving those engines a reason to keep recommending you after the first question. No email required to get your diagnosis.
Sources: Search Engine Journal, "62% Of AI Brand Recommendations Vanish After One Buyer Question - New Clovion Data," by Greg Jarboe, July 7, 2026 (searchenginejournal.com).