How AI Assistants Decide Which Brands To Recommend

Refusals matter. A report that only contains successes is either describing a suspiciously easy month or omitting the parts that did not work, and the omitted parts are usually where the useful information is.

The result is a content programme aimed at guesses. Sometimes it works by accident. Usually it produces pages nobody retrieves, and the diagnosis that would have directed the effort correctly costs a fraction of what the content did.

A final error deserves separate mention because it undoes good work rather than merely wasting effort. Teams that get an early result frequently conclude they have found the mechanism and generalise from one change. A directory correction coincides with a mention appearing, and directory corrections become the strategy, when the actual cause was a rewritten page indexed the same week.

Be wary of proposals where the largest line is content production. It is the easiest work to scale, the easiest to bill and the least likely to be the constraint, particularly before a baseline exists. A proposal weighted toward diagnosis, technical fixes and third party corrections is usually cheaper and almost always sequenced better.

Starting With Content The most common and the most expensive. A brand decides to take this seriously and commissions twenty articles, without knowing which questions matter, which assistants answer them badly, or which sources those answers are built from.

The sustainable version is small and continuous: the prompt set run monthly, listings checked quarterly, a handful of pages updated rather than a burst of new ones, and someone who owns it. That costs less over a year than the three month push and holds its ground. ai visibility agency

In most categories the pages that generate answers are review platforms, directories, forum threads, comparison articles and trade publications. Being absent or wrong on those explains far more absences than anything on a brand's own site, and correcting a listing costs an afternoon.

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.

Good versions read like this: mention rate on evaluation prompts rose from two in fifteen to six in fifteen, which we attribute to the three directory corrections completed in week two, though a competitor also stopped publishing during the same period.

A Claim About Causation, Stated as a Claim The most valuable paragraph in the report is the one that says what moved and why the agency believes their work caused it, phrased as a judgement rather than a fact.

Buying a Score Instead of Evidence A monthly number that rises is easy to present and impossible to audit. The vendor controls the number and the prompt set behind it, and a client has no way to distinguish real improvement from a methodology change.

The risk is scope drift into activity that is easy to report and hard to value. The protection is to have the retainer specify countable units: prompt set runs per month, listings audited, corrections submitted, pages published or rewritten, outreach attempts made.

What Ranking Does and Does Not Buy You Ranking still helps, because the retrieval step usually starts with a search. But it buys far less than people assume. Ahrefs examined 15,000 long-tail prompts across four assistants in July 2025 and found roughly 80 percent of cited pages did not rank for the original query at all, with about 12 percent in the top ten.

Retrieval behaviour changes, competitors keep publishing, listings go stale, product details change and reviews accumulate. A position secured once is not held without maintenance, which is the same lesson search taught over twenty years and which is being relearned rather than transferred.

We also know the picture is unstable. Retrieval strategies are revised without announcement, and a method that explained answers well six months ago may explain them poorly today. Anyone selling certainty here is selling something they do not have.

Consistency Matters More Than Anywhere Else Local identity resolution depends on the business details agreeing across a long tail of directories, many of which nobody has looked at in years. Old addresses, disconnected numbers and previous trading names sit in these places indefinitely.

Why Ranking Stopped Guaranteeing Visibility The assumption underneath two decades of search marketing was that position and visibility were the same thing. Retrieval based answering breaks that link, because the pages a model reads to compose an answer are not necessarily the pages that rank for the question.

The practical conclusion is unexciting and reliable. Do the work that pays off under multiple scenarios, keep measuring, and treat any strategy that requires one channel's terms to stay fixed as a bet rather than a plan. ai visibility agency