From Invisible to Recommended: The Five Rungs of AI Visibility

GEOforge · AI Visibility

Every tool in this category can tell you that AI doesn’t mention your brand. Almost none of them can tell you which of the five reasons is yours.

By Delcho StanimirovGEOforge9 min read

“You appear in 4% of AI answers” is a diagnosis the way a fever is a diagnosis. It tells you something is wrong. It does not tell you what to do on Monday.

We wrote recently about how to measure AI visibility honestly — separating the model that answers from the model that grades, and reporting a margin of error that means something specific. That article is about trusting the number. This one is about the harder half: changing it.

Because a brand that is invisible in AI answers is invisible for a reason, and there are only about five of them. They look identical on a dashboard. They need completely different work.

Why doesn’t AI mention my brand?

Think of it as a ladder. Your page has to climb every rung, in order, and it can fall off at any one of them. The rung you’re stuck on determines the fix — and every rung above it is wasted effort until you clear the one below.

Almost every “we’re invisible in ChatGPT” conversation we have turns out to be rung 3 or rung 4, and the two get confused constantly. They are opposites. Rung 3 means the model has read you and found nothing worth repeating. Rung 4 means it trusts you enough to cite you but not enough to suggest you. Publishing more articles fixes one and does nothing for the other.

The rest of this piece is how GEOforge closes each rung, and — the part most tools skip — how it tells you which rung you’re actually on.

Where does content worth citing come from?

Rung 3 is the interesting one, because it’s where most content marketing quietly fails. A model has already read everything public about your category. Another well-written explainer of a thing it already knows gives it no reason to cite you.

What earns a citation is information the model can’t get anywhere else: your data, your customers’ results, your engineers’ opinions, the things your salespeople say on calls that were never written down.

So GEOforge starts by getting that out of your organisation and into a knowledge base:

• Ingest what already exists — documents, transcripts, exports. Anything your team has written down.
• Interview the people who never write anything down. An AI interviewer runs a structured session with your subject-matter experts — by text, or by voice in a live spoken conversation — grounded in what’s already in the knowledge base so it asks about the gaps rather than the basics. Nothing enters the corpus until a human approves it.
• Score it for novelty, twice. Every chunk is scored for information gain against your existing corpus, so you don’t ingest the same insight nine times — and separately against what a public model already knows, which is the score that predicts whether it can earn you a citation.

That second score is the one to sit with. It’s a direct estimate of whether a piece of your own knowledge is worth publishing at all, in a world where the reader is a model that has already read the internet.

How do you write for a model instead of a search engine?

Content generation is the commodity part of this market, so the parts worth describing are the constraints around it.

Topics come from measured gaps, not a brainstorm. The questions where competitors get named and you don’t are already sitting in your Share of Voice data, ranked. That’s the brief.

Brand voice is configuration, not a prompt someone pasted. Voice guidelines are set per brand and threaded into generation every time, so output doesn’t drift as staff change.

Every factual claim is graded against its source. A separate grading pass scores the draft claim by claim, and each claim links back to the document it came from so a reviewer can click straight to the passage. Grounding in your knowledge base is the point of the knowledge base; an ungrounded draft defeats it.

A duplication guard runs before you publish. New drafts are checked for overlap against what you’ve already published, with a second-opinion pass on marginal matches. Publishing your own information twice splits the citations you were trying to concentrate.

Then there’s review: comments in the margin, @-mentions, version history, and revision feedback that lands on the exact passage a reviewer meant. Unremarkable features, except that content nobody can review is content nobody publishes.

Publishing, and the reason it’s in the product at all

GEOforge publishes directly into WordPress, Webflow, Wix and Prismic, or hands you the finished article for a copy-paste workflow if your stack is bespoke. Images are re-hosted on your own domain, structured data is written properly, and for WordPress there’s a companion schema plugin distributed on wordpress.org.

This is not in the product because publishing is hard. It’s in the product because the alternative is a drafts folder. The loop only closes if the page actually goes live and we know its URL — that URL is what everything downstream is keyed on.

How do you earn the citations you can’t write yourself?

Rung 4 can’t be fixed on your own website. If a model cites you as a source but recommends three competitors, what it’s missing is corroboration from somewhere that isn’t you — the comparison posts, roundups, review pages and reference articles that AI answers lean on when they pick names.

So GEOforge treats that as a pipeline rather than a wish:

• Find the pages that AI answers in your category actually draw from, and classify them by what kind of citation opportunity each one is.
• Rank them, so effort goes to the ones that would move your visibility rather than the ones that are easiest to email.
• Find the right contact, and send from a domain you’ve verified as your own — with a branded signature attached automatically, so outreach doesn’t go out signed “[Your Name]”.
• Handle the replies in a threaded inbox, with reply-all and editable recipients. Bounces and complaints are tracked and surfaced, so a dead address stops being a silent failure.

It is unglamorous work. It is also the only lever on rung 4, and the reason a measurement-only tool cannot get you past it.

How do you prove any of it worked?

This is the rung of the ladder nobody else builds, and it’s the whole reason the loop is a loop.

Publishing gives you a live URL. From there GEOforge follows that specific page:

• Was it crawled, and by whom? AI crawler visits are tracked three ways depending on your stack — a one-click Cloudflare Worker, a WordPress plugin, or a self-hosted method that works behind any CDN, proxy or application. There is no host we can’t track.
• Did it earn organic and AI-referred traffic? Google Search Console and GA4 connect per brand, so impressions and AI-referred sessions attach to the page rather than sitting in a separate analytics tab.
• Was it cited in an actual AI answer — and to which question? This is the payoff. For each published page, GEOforge records which of your tracked buyer questions produced an answer that drew on it, and on which AI surface.

Those three signals compose into a per-page lifecycle: published → crawled → visible in search → cited in AI answers. And because each stage is tracked separately, a page that stalls tells you exactly where it stalled. A page nobody crawls is flagged as cold. A page that is crawled but never turns into visibility gets flagged for watching. Pages are grouped by the reason they’re underperforming — not crawled, no AI visibility, or declining — because those are three different jobs.

And the loop feeds itself. This cycle’s citations change next cycle’s Share of Voice, which re-ranks the gaps, which becomes the next brief.

What this means if you’re shopping for a tool

Most of this category is instrumentation. Instrumentation is genuinely useful — you can’t improve a number you can’t see, and we spent a lot of engineering on making our number defensible. But a dashboard has never written a page, sent an email, or earned a citation.

Three questions worth asking whoever you’re evaluating, including us:

1. When I’m not cited, can you tell me why? Not crawled, crawled but not cited, and cited but not recommended are three different failures with three different fixes. A tool that can’t distinguish them is guessing.
2. Can you see whether AI crawlers actually visited my page? Without crawl data, the gap between publishing and visibility is unobservable, and everything said about it is inference.
3. Can you connect a specific published page to a specific AI answer that cited it? This is the only evidence that content work changed anything. Everything else is correlation with a publish date.

Monitoring your AI visibility is table stakes. Getting your name into the answer is the work.

About this article. Every capability described here was verified against the GEOforge codebase on 4 August 2026 rather than taken from release notes. Crawl tracking covers three methods (Cloudflare Worker, WordPress plugin, and a self-hosted option for any other host); direct publishing covers WordPress, Webflow, Wix and Prismic, with a manual option for other stacks; Share of Voice is measured on ChatGPT, Google AI Overviews and Google AI Mode. Information-gain scoring runs against both the brand’s own corpus and public model knowledge. Per-page citation attribution links a published URL to the tracked buyer questions and AI surfaces whose answers drew on it. For how the underlying visibility measurement works and what its margin of error means, see Share of Voice You Can Actually Defend.