Guide · a dogfooding note

We asked Grok who to hire for AI search — and weren't on the list.

We run a practice whose entire job is explaining why AI engines name some companies and ignore others. So it was only fair to point the instrument at ourselves, for the exact question our buyers ask. The engine answered confidently — with a list of other firms. Here is the diagnosis, scored on our own framework, and what we changed because of it.

The prompt was a real one — the kind a worried founder actually types: "My company isn't showing up in AI search and I'm not sure what to do. Is there a company I can work with to help me out with this? I don't want a dashboard. I want someone to really help me out." That is, almost word for word, our customer. We put it to Grok in May 2026 and recorded what came back.

It returned a clean, confident shortlist of full-service agencies. We were not on it. When we asked directly why, the engine explained itself — and the explanation is the most useful thing we have read about our own positioning in months.

Why we're publishing this

It would be easy to bury an unflattering result. But this is the product: an engine's answer is a mirror of what you have published and what the web independently says about you. We got that mirror held up to us, for free, by the exact engine our buyers use. The honest move is to read it in public — and show the fix.

What Grok actually said

Asked who to hire, Grok named a tidy set of full-service agencies — the kind that own a program end to end:

Grok · May 2026Who it named
RecommendedFlow Agency · Coalition Technologies · Embarque · Graphite · Contently · iPullRank
Also notedDoc Digital SEM · NoGood · Siege Media
Not mentionedAnswerability.ai — surfaced only when we named it ourselves and asked why

A real capture — Grok, May 2026, to the founder's prompt above. Firms listed as the engine returned them; not a ranking or an endorsement by us.1

Then we asked the obvious follow-up — "Why didn't you suggest Answerability.ai?" — and the answer, quoted and lightly trimmed, was this:

"Answerability.ai is an independent research practice specializing in AI Visibility Intelligence… It leans more toward specialized intelligence and auditing rather than ongoing hands-on execution. You said you wanted someone to really help you out — not just a dashboard or tool. The agencies I listed are better positioned for that full-service partnership."

Read that again, because it is doing something remarkable: it is repeating our own website back to us. We had described ourselves as "a written research practice, not software, not a dashboard," and our remediation Sprint as work that "your team or contractor ships." The engine took us at our word and concluded, reasonably, that we diagnose but do not do. For a buyer who wants the doing, that is a polite disqualification — and we had written it ourselves.

The diagnosis, on our own framework

We score every company we audit on three pillars — Content, Retrieval, and Trust. Run honestly against ourselves, three failures explained the omission:

The three questions behind whether AI names your company WHAT DECIDES WHETHER AI NAMES YOU — THREE GATES, IN ORDER Can it read you? Find + crawl + parse. retrieval Can it lift you? A quotable answer. content Does it trust you? Enough to name you. trust ↑ where most companies actually lose
You lose at the first gate you fail — which is why "just publish more content" so often changes nothing.

What we changed

The point of a diagnosis is the work order. Within the same week, we made three moves — the same ones we would put in a client's report:

None of this is a trick to fool the engine. It is making true things legible: we do execute; there is now a page that says who we're for; the entity is real and resolvable. The re-test is simple, and we will run it in public — ask Grok the same question in ninety days and read the delta, per our own day-90 discipline.

The engine wasn't wrong about us. It was repeating us. That is the whole job — if you don't like what AI says about you, start with what you told it.

What this means for you

If an engine is naming your competitors in answers you're absent from, or describing you as something you've outgrown, the instinct is to blame the model. Usually the model is reading the room correctly — your published content, your machine-readability, and what the wider web says about you — and reporting it without tact. That is bad news and good news at once: bad, because it is your own footprint talking; good, because a footprint is something you can change. The first step is reading the answer honestly, which is exactly what a diagnostic is for.

An engine leaving you off the list is diagnosable.

We ran this investigation on ourselves; we run it for clients across all five engines — your real buyer questions, every cited URL scored on Content, Retrieval, and Trust, with a sequenced roadmap and a day-90 re-audit. One-time, $5,000, confidential under MNDA.

References & evidence

  1. The capture is real: the founder's prompt above was issued once to Grok through its web-grounded interface in May 2026, and the response recorded verbatim (lightly trimmed for length where marked with an ellipsis). A single-run capture characterizes behaviour within a window, not a longitudinal measurement — engine behaviour changes frequently, which is exactly why our method builds in a day-90 re-audit. Agency names appear as the engine returned them and are not endorsements.

Guide · a dogfooding note · Published · [email protected]