How AI Assistants Decide Which Brands To Recommend

The change worth making is editorial direction. Stop commissioning new pages whose entire value is a fact a summary can state, and redirect that effort toward comparison, judgement, original data and anything requiring a transaction. Keep the existing pages, keep them current, and structure them to be quoted.

This matters more than any subtlety about model training. It means recommendations are built largely from pages that exist right now, which is why a page published this month can influence an answer this month, and why a brand absent from the retrievable web is absent from the answer regardless of how well known it is offline.

What to Do About llms.txt and Similar Files Proposals for machine readable files aimed specifically at language model consumers appear periodically. Adoption is inconsistent and support varies by provider, so treat these as low cost and speculative rather than as a requirement.

Then segment by query type. If the decline concentrates in informational and definitional queries while transactional and comparison queries hold, the cause is almost certainly something above you answering the question. If the decline is even across every query type, look elsewhere, because that is a different problem.

Structured data attracts a particular kind of over-investment. Teams implement a dozen schema types, validate them all, and conclude the job is done, having spent most of their effort on markup that changes nothing about how a machine understands the business.

The Broader Lesson Every stage of this has punished the same thing, which is dependence on a single channel whose terms you do not set. Featured snippets did it, each core update did it, and this is doing it again with more force.

A quick way to find contradictions is to write out your key facts on one sheet, taken from your structured data, then check that sheet against your about page, your main directory listing and your marketplace account. Doing it manually feels crude and it surfaces the conflicts that validators never flag, because a validator checks syntax rather than whether your founding year matches the one you published elsewhere.

What Has Not Changed It is worth being clear about the continuities, because the change is regularly oversold. Organic search still delivers the larger share of traffic for most businesses. Crawlable, fast, well structured sites still win. Content that genuinely answers a question still outperforms content that does not.

In that setting your ranking is one input among several to a retrieval step, and often not a decisive one. Ahrefs found in July 2025, generative engine optimization across 15,000 long-tail prompts, that around 80 percent of cited pages did not rank for the original query at all, with about 12 percent in the top ten.

One practical note on prioritising outreach. Sort your citation list by frequency and start at the top, not at the most prestigious name. A directory that appears in half your category's answers is worth more than a publication that impresses your board and has never been cited once. This is the point at which visibility work and conventional public relations objectives diverge, and it is worth saying out loud before the two budgets start competing.

A simple system beats a campaign. Ask every satisfied customer, at the point where they have just been satisfied rather than a month later. Make it one click. Respond to everything, briefly and without defensiveness.

The Mistake That Undoes Everything Markup is a claim, not evidence. Structured data asserting that you own a profile only helps when that profile exists and points back at you. Markup naming an author only helps when the author can be found elsewhere.

What We Genuinely Do Not Know Several things are worth admitting rather than papering over. We do not know how the systems weight their signals against each other. We do not know how much residual influence training data has once retrieval is involved. We cannot reliably distinguish a change in your visibility from a change in the model's behaviour.

The Mechanism Most Answers Now Use The common architecture is retrieval augmented. Your question triggers one or more searches, a set of pages is fetched and read, and the model writes an answer grounded in what it just read. Citations, where shown, point at those fetched pages.

None of them are harmful. They just consume implementation and maintenance time that would achieve more if spent making the Organization markup accurate everywhere, or correcting the directory listing that has your old address on it.

Give journalists and analysts accurate material to work from, in a form they can use without rewriting. Where an independent comparison exists and gets your details wrong, a polite factual correction is accepted far more often than people expect, because publishers generally do not want to be wrong.

Read alongside the first displacement, the picture is consistent: the top of the list is worth less than it was on the results page, and worth considerably less again in a channel that does not use lists.