How to choose an AI visibility (GEO) tool

Measuring answers, not rankings.

A GEO — generative engine optimisation — tool asks AI assistants the questions your buyers ask and records what comes back: whether you were mentioned, whether you were cited as a source, and who was cited instead. That's a different measurement from search rank, because the assistant synthesises one answer rather than listing ten links.

The better tools go further than measurement: they look at why a page was passed over, turn that into work someone can do, and check the work landed. What none of them can do is control the engines. Treat any promise of guaranteed citations as a red flag.

It's also worth being clear about the levers. Machine-readability work — structured data, clean markup, an llms.txt index — helps engines parse your pages and attribute them to the right entity, and crawler behaviour varies between OpenAI, Anthropic, Perplexity and Google. But the strongest levers remain answer-shaped content, genuine topic coverage, and being cited by sources the engines already trust.

Five capabilities worth demanding — and where we fit.

Tracking across multiple engines, on a schedule

One-off checks tell you nothing. Ask how many assistants are queried (ChatGPT, Gemini, Claude, Perplexity, Copilot and Google AI Overviews behave differently from one another), how often, and whether results are stored as a trend you can look back through. A single score with no history can't distinguish a real change from a bad day.

GEOquoted: Weekly runs across the major assistants, stored as a per-prompt history with Share of Voice against competitors you nominate.

Evidence-backed diagnosis, not a vanity score

A number out of 100 isn't actionable. Ask what the tool can tell you about a specific page: what the engines cited instead, which parts of your page are hard to parse, where coverage is thin. The useful output is a named finding with the evidence attached, not a grade.

GEOquoted: Findings are tied to a URL and quote what we saw on the page and in the answer, so you can check our reasoning.

A workflow, so findings actually get shipped

Most audits die in a PDF. Ask whether findings become assignable work with an owner and a status, and whether the tool drafts the change itself — the JSON-LD block, the FAQ section, the content brief — or just describes it.

GEOquoted: Every finding becomes a job with the artifact drafted, moving through To do → In progress → Completed, assignable to whoever owns the page.

Independent verification that a fix went live

Marking work as done is not the same as the work being done. Ask whether the tool re-fetches the live URL and confirms the change is present before it counts as complete, or whether it takes your word for it.

GEOquoted: When someone marks a job done, we fetch the URL and check for the change; jobs sit in 'awaiting verification' until they pass, and failures come back with what we found instead.

Before/after proof tied to a specific change

Ask how the tool attributes improvement. Comparing this month to last month across everything you did proves nothing in particular. You want a window anchored to a dated, verified change.

GEOquoted: Each verified fix anchors a before/after window comparing Share of Voice, citations and AI-referred sessions either side of that date, shareable as a snapshot.

Four reasons to walk away from us.

You want guaranteed rankings or citations. Nobody can offer that, and any tool that implies otherwise is selling something the engines don't sell.

You need traditional SEO tooling — rank tracking, backlink indexes, site crawls at enterprise scale. We assume you already have that and sit alongside it.

You have no one to ship changes. GEOquoted drafts the fix, but someone still has to publish it. Without that, you'll get a well-evidenced list and no movement.

You want a single number for a board deck and nothing else. Our output is a working queue; the reporting is a by-product of doing the work.