The AI Visibility Brief: July 6-12, 2026

THIS WEEK

• 92% of B2B buyers say AI has shaped their vendor shortlist, and only 7% notice a vendor by name
• Deleting a page does not remove it from AI answers
• A third of AI answers built from open web search cited a source experts flagged as untrustworthy or off-topic

01

92% of B2B buyers say AI has shaped their vendor shortlist, and only 7% notice a vendor by name

Semrush surveyed 643 US B2B professionals in March and April on how AI plays into choosing vendors, and published the results July 8. Among the 519 who use AI at work, 92% said AI tools have influenced which vendors made their shortlist. Recognizing a name barely mattered: just 7% pointed to brand recognition as the thing that makes a vendor stand out in an AI answer. What stood out instead was a close match between the vendor and the buyer's specific need.

We think this matters because a B2B shortlist used to come from searches, referrals and names the buyer already knew. Instead, almost all of these buyers now say AI tools have a hand in it. Buyers are choosing vendors that match their specific need.

Businesses should act on this by publishing the specific situations they serve best in plain buyer language, on their own website, on LinkedIn, or in the community forums where the category gets discussed. When a buyer asks for the best payroll provider for a 20-person accounting firm, the AI models are pulling from websites that describe serving exactly that kind of firm.

02

Deleting a page does not remove it from AI answers

Otterly published an experiment July 10 on what happens to AI citations when the cited page gets deleted. One brand took down 15 of its comparison pages on known dates while Otterly watched citations across seven AI search engines every day, from before each removal through the months that followed. The citations did not stop. Every one of the 15 pages kept appearing in answers after it was gone, some for weeks, one batch for more than eight months, and Gemini alone produced 31% of these ghost citations, answers built on pages that no longer exist.

We think this matters because these engines work from a memory of the web, and that memory holds on long after a page is gone. If a page says something wrong or outdated about a business, deleting it does not fix the answer. The answer keeps citing the page, and the link behind it now leads nowhere.

Businesses should act on this by fixing a page that says something wrong rather than removing it, either updating the content where it stands or pointing the old address to a current page, since both keep something accurate at the end of a citation that is going to keep showing up either way.

03

A third of AI answers built from open web search cited a source experts flagged as untrustworthy or off-topic

A team at the University of Iceland tested a public EU-information service ahead of its launch and posted the results to arXiv July 6. The test compared answers the system built from the open web against answers built from a small vetted set of sources. Five experts graded 449 AI-written answers along with every source those answers cited. When the system searched the open web, 35% of its answers used at least one source the experts considered untrustworthy or irrelevant. When it worked from the vetted collection instead, complaints were rare and only ever about a source being out of date.

We think this matters because these tools sometimes pull from untrustworthy websites, and the reader gets no warning when they do. The answers are only as honest as the pages they are built from. Putting out accurate, honest material is how a business protects what gets said about it.

Businesses should act on this by writing the pages they control the way reference material gets written: facts stated plainly, claims that name where they came from, a real author on the page, and a date showing it is kept current. Those are the things that build authority and improve trust in a source.

Thanks for reading this week’s newsletter.
Sincerely,
Brett