Carolyn Shelby in Search Engine Journal says AI search returns outdated brand facts because old pages still match the question

In a column published in Search Engine Journal on September 7, principal consultant Carolyn Shelby says the biggest brand risk in AI search is not too little content but too many versions of the truth still sitting on the brand's own properties.

Shelby writes that traditional search could rank several dated pages side by side and let the user pick the current one. AI search products no longer work like that. They retrieve sources and use them to build a single answer, and the answer looks settled even when the evidence behind it is not.

Why the old page keeps winning

Shelby's mechanism is about retrieval. The user's prompt supplies the frame and much of the vocabulary the system searches with. When the question contains an assumption that is out of date, the retrieval favours pages that use that same wording.

Her worked example is a leadership title. Someone asks an AI product who the CEO of a company is. The company no longer has a CEO in that shape, and the current senior leader carries a title like SVP and GM. Old bios, press releases and conference profiles all name former CEOs by that exact title. The new leadership page uses the new language. The model, Shelby writes, "cannot cite a source it did not retrieve," so the current title never enters the answer and the old name wins.

She names four executives in one real case, each of them a past holder of the title, none of them the person actually running the business now.

What Shelby says to do about it

Shelby's prescription is bridge content. A brand needs pages that explicitly link the obsolete vocabulary a prompt is likely to use to the current reality: a sentence that names the old title and the new title in the same paragraph, so retrieval on the old question can find a page that also carries the new answer. Publishing a new leadership page on its own does not do this, because a page written in the current vocabulary will not be found by a question written in the old.

The wider argument is that AI search has made content governance an SEO problem. Every outdated PDF, executive bio, product description, or partner-page mention is now a candidate answer, and the brand is at the mercy of whichever one the retrieval prompt happens to match. Shelby's audit target is the evidence chain, not the surface content: the set of pages the model might reach for, ordered by how well they match the questions a user is likely to ask.

The specifics beyond that are hers, in her own words on Search Engine Journal, and worth reading against your own site rather than paraphrased. The action she asks of a brand is a content-inventory pass through the site with the retrieval question in mind: what does the model find first when it looks for the fact your buyer is asking about, and does that page still describe the company you run today. Fresh publishing is not a substitute for pulling or updating the old pages that will otherwise keep answering the question first.

Source

• Your Biggest AI Search Risk Is Conflicting Information About Your Brand by Carolyn Shelby, Search Engine Journal, September 7, 2026.