Avouch
Menu

The machines had this firm wrong. Watch them change their answers.

There’s a difference between showing up and being recommended.

Verified client result · anonymised
FIG. 01, Before / after

User prompt: What does [the client] do?

ChatGPT

[The client] is a company that specializes in building and managing remote engineering teams.

Wrong, and not close. Half of all model-memory answers about the company were this kind of guess.

User prompt: Who are the best providers in [the category]?

Gemini

Best for mid-market and cost transparency: [the client]1, a standout option if you want to avoid hidden fees.

★ NON-BRANDED PROMPT · NAMED AMONG 5

The two questions were asked nine weeks apart, and both are on the record. One is an engine guessing from memory; the other has read the client's site and answers accordingly. The client's name and category are redacted.

Cited in ChatGPT, Claude, Gemini, Perplexity, Grok, Brave, Google AIO, Copilot.

Real, but unknown

A B2B firm with genuine expertise, up against rivals many times its size. The expertise was real; the engines had simply never heard of the company.

In May 2026 we ran the baseline audit. Half of the engines' from-memory answers about the company were hallucinations: a dev-team marketplace one moment, someone else's project-management tool the next.

Two months in, pure model memory still hadn't budged. A July probe put 68 non-branded buyer questions to six models: the client was named zero times, and the category leader up to 34. Memory moves last, and that is the pre-registered December target.

AI can't recommend what it doesn't know.

The diagnosis is an entity + naming problem, not distribution. Engines were already reading the client's site. They answered branded questions correctly 99.1% of the time. They just never volunteered the name.

Keep
  • Real expertise, branded questions answered correctly 99.1% of the time.
+
Start
  • A machine-readable identity, content where citations already cluster, and vocabulary the engines could adopt.
Stop
  • Treating entity, content and measurement as separate projects instead of one ordered sequence.

Entity first

A machine-readable identity on every page: schema, Wikidata, Companies House anchors, one brand string. Put first because entity effects lag by months.

Content where citations happen

Cost and comparison pages earn 8 of the client's 10 most-cited URLs, a comparison hub, cost guides, an interactive calculator, and location pages gated on verification.

Vocabulary the engines adopt

The client published its own named frameworks. By April, Google's query log showed fully-formed questions about them ranking at position ~1.

Measurement that survives an audit

759 prompts, seven engines, 9,592 answers. 78 claims adversarially checked: 66 confirmed, 12 corrected, none refuted. The corrections were published.

FIG. 02, What moved
22%of tracked AI answers cite the client's domain, up from 16%

From 24th to 6th most-cited domain in Avouch's own sweep, May to July, and the most-cited domain of all, branded or not, across all tracked AirOps prompts, three weeks running.

552%Google clicks, February to June 2026 (Ahrefs-connected GSC).

June's organic-search opportunity creation ran 230% above the 2025 monthly average, and the first half of July had already beaten the strongest full month of 2025. One buyer clicked a ChatGPT citation of one of the client's comparison pages and was a pipeline deal 51 seconds after the form went in; the meeting is now booked.

Verified client result · anonymised

Cited everywhere, named almost nowhere

On the same July questions, the client is still last of the tracked brands on being named: about 2%, against roughly 30% for the category leaders. Engines lean on the client's pages as a source, then recommend somebody else.

That gap is the next phase, and we've pre-registered the December targets rather than marking our own homework. If the naming number doesn't move, you'll see that too.

Go where the citations already are

The client's own pages are already the most-cited in the category. What they can't out-run yet is domain authority: the category's leaders sit far higher, and the client has climbed roughly ten points from a low base. Closing that gap isn't more content on their own site. It's presence on the domains AI engines already trust and quote directly.

  1. PR

    Earned press links are the direct lever on the domain-authority gap itself, the one number the client's own site can't move alone.
  2. YouTube

    youtube.com is already the single most-cited domain in Google's AI Overviews for this category, ahead of every competitor's own site.
  3. Reddit

    Buyers already find the client through Reddit threads. A citation there doesn't need the client's own domain authority at all, it borrows Reddit's.
  4. Review platforms

    A reconciled, growing review count gives engines a neutral comparison source to cite, instead of guessing at who's best.

We went where the data said we needed to be for the pages. This is the same logic, applied to the domains the pages can't reach alone.

Most-cited in two months, named almost nowhere

The client's expertise never changed. The work made it legible to the engines its buyers ask first, measured weekly, with every correction published.

AI can't recommend what it doesn't know.

Check if AI recommends you

Start free. No commitment.