The Honest Limits of AI Visibility

Every new discovery channel breeds the same industry twice: first the people doing real work, then the people selling certainty. AI visibility is young enough that both are calling your practice this year, and they sound similar on the phone. This page is the vaccine.

Start with the claim that settles most conversations:

Nobody can guarantee an AI recommendation. Nobody. Assistants don't publish their reasoning; they show sources, not rankings — you can see who they cited, never why those names won. Answers vary by phrasing, day, user, and model version. There is no dial anyone can turn that makes an assistant name you. What exists is the record assistants read — and moving every checkable factor in your favour moves the odds. The odds are all there is, and anyone selling more than odds is selling smoke.

The snake-oil patterns

The pitches follow recognisable shapes. Any one of these ends the meeting:

Guaranteed placement. "We'll get you into ChatGPT's answers" — see above. The honest version of this sentence doesn't exist.

The proprietary method. "Our AI-optimisation algorithm" for a system whose inputs are public and boring: legible pages, consistent records, real reviews, citable substance. There is no secret layer to have a proprietary method about.

The urgency lever. "Your competitors are already optimising — every week you wait, the answers calcify." Answers refresh with the record; nothing calcifies. Early effort is cheap because the field is empty, not because a door is closing.

The unfalsifiable report. Monthly dashboards of invented metrics — "AI visibility score up 12%" — that connect to no observable answer. The only honest measurements are the ones you can replicate yourself: the monthly ritual, and audits whose every check you can inspect.

What honest work looks like

Real AI-visibility work is auditable by construction, because it lives in the checkable record: it fixed the pages that didn't render, stated the facts that were missing, reconciled the listings that disagreed, and can show you each one. It measures itself against logged assistant answers over months, and it tells you when the constraint isn't this at all. That's the standard this library holds its own Get Found by AI Audit to — every factor scored is a factor you can verify — and it's the standard to hold any vendor to.

And one limit that belongs to you rather than the vendors: this channel rides on substance. A thin practice with a perfect record is still thin — the assistants are cautious librarians, not magicians, and what they ultimately index is whether you're verifiably good. The odds-moving work is real; it just can't outrun the practice itself.

Questions practices actually ask

So is paying for AI-visibility help ever legitimate? Yes — as bounded, itemised, checkable work: a legibility fix-list executed, a listings reconciliation completed, an audit with inspectable checks. Pay for artifacts you can verify, never for influence over answers.

An agency showed me a screenshot where ChatGPT recommends their client. Proof? A screenshot is one answer on one day to one phrasing. Ask instead for the record work they did and the client's answer-log over months — the honest version of the same evidence.

What results are realistic? From a weak record: appearing at all within a season of fixing legibility and agreement, then named-with-specifics as reviews and substance accumulate. From a strong record: defending a position mostly by staying current. Anyone quoting precise percentages is reporting decoration.

Why does this page undercut the audit this site sells? Because the audit only makes sense inside these limits: it scores the record, ranks the fixes, and promises odds — nothing else. If this page's standard ever stops being the audit's standard, trust neither.

---

Part of Get Found by AI — the deep dive on the second front door.