AI Visibility, explained

Why AI visibility matters

A growing share of "best business in [city]" queries no longer happen on Google's search results page — they happen inside an AI chat. ChatGPT users ask "what's a good plumber in Sacramento?". Gemini users ask the same in their Google app. The AI returns a list. If your prospect's business isn't on that list, they're invisible to the fastest-growing local-search channel of 2026, no matter where they rank in classic SEO.

AI Visibility is the section of the dossier that turns this into something measurable.

How the probe works

For each dossier, we send three customer-style prompts to two AI engines:

• ChatGPT via OpenAI's gpt-4o-mini-search-preview — a model with built-in real-time web search.
• Gemini via Google's gemini-2.5-flash with Google Search grounding enabled.

The three prompts are admin-tunable, but the defaults are:

• "Best {category} in {city}."
• "{category} near {city}."
• "Find a reliable {category} in {city}."

Each prompt asks the AI to return up to 5 businesses it knows about, formatted as { "name": "...", "website": "..." }. We use JSON mode (when supported) so parsing is deterministic.

That's 3 prompts × 2 engines = 6 total probes per dossier.

How "cited" is detected

We use a strict URL-only rule. The prospect counts as cited only if the prospect's domain appears in one of the businesses' website fields. Name-only matches don't count, because AI engines often invent plausible-sounding business names that happen to match a real prospect's name by coincidence — that's not a real citation.

Since the visible list and the citation check use the exact same parsed entries, you'll always be able to see why a probe was marked cited: the prospect's URL will be in the displayed list, highlighted in accent color.

The score

The AI Visibility Score is simply (probes cited / total probes) × 100:

• Cited in all 6 probes → 100/100
• Cited in 3 of 6 → 50/100
• Cited in 0 of 6 → 0/100

It's deliberately a coarse score. Three engines, three queries each, hyperlocal context — the signal is "do AIs know this business or not?", not a precise market share.

Why ChatGPT and Gemini might disagree

They will. ChatGPT's web search returns one set of pages; Gemini's grounding hits a different result mix. Different prompts return different lists. This is fine — the contrast itself is informative on a call: "Gemini knows you, ChatGPT doesn't. Different audience exposure for each engine."

Why the prospect is sometimes empty

When a hyperlocal business with a small or new web presence isn't cited by either engine across all 6 probes, the score is 0. That's not a bug — it's the truth. Use it on the call:

"When customers ask AI for a {category} in {city}, neither ChatGPT nor Gemini names you. They name {competitor 1, competitor 2}. The fix isn't another rank-tracking dashboard — it's the technical and content work that gets you into AI training and grounding sources."

Reruns and prompt customization

If a probe returns oddly empty results or an outlier list, you can rerun just the AI Visibility section using the Rerun button on that card. Each dossier has 3 reruns budgeted for AI Visibility specifically (the most expensive source). Reruns refresh the prompts from the latest defaults, so any prompt updates in the admin panel take effect immediately.

What we deliberately don't do

• Perplexity, Claude, Google AI Overviews. We tested these. Results were either redundant with ChatGPT/Gemini or unreliable enough to not be worth the cost. We may add them later as on/off knobs.
• Aggregator and directory mentions. The prompt explicitly tells AIs to skip Yelp, Yellow Pages, BBB, Angi, etc. We want what the AI says directly, not what its scraping pile-up of directories returns.
• Ranking position within the AI response. The score doesn't differentiate between #1 and #5 — both count as cited. The signal is binary: visible or not.