How to measure AI visibility without hiding the gaps

Your leadership team wants a number. Sixty-two percent AI visibility, up from last quarter. The board deck needs a trend line, and a single percentage is easier to defend than a spreadsheet of buyer questions.

Someone on the team manually opens ChatGPT once a week, types a buyer question, and reports whether the brand appeared. A vendor dashboard may already show a blended score stable enough for a slide. That blended number cannot tell you which buyer questions include your brand, which omit you, or which recommend a competitor instead — so a healthy average can hide pipeline risk on evaluation questions.

If you cannot see coverage per question, you are measuring AI visibility without seeing where you win or lose. The fix is per-question presence — which buyer questions mention your brand, broken down by model. ShareOfAsk calls this Question Presence, on the Question Presence coverage map.

Why one visibility score hides where you win or lose

Aggregate visibility tells you how often you appear across a tracked set. Coverage shape tells you which questions drive or depress that average. The two can look healthy together while telling opposite stories about pipeline risk.

A buyer asks Claude which vendors handle SOC 2 for a 50-person team. Your brand is absent. The same week, a dashboard reports sixty-five percent visibility because you appear on “what is [category]?” and “best tools for beginners.” The average is fine. The evaluation question that filters a shortlist is not.

A blended score summarizes presence; it does not map it. Portfolio KPIs such as Visibility, Share of Voice, and Consideration Share roll up competitive outcomes for a snapshot period. Question Presence answers where — the buyer questions behind those rollups. See how they work together on Visibility, Share of Voice, and Consideration Share.

“What is a good AI visibility score?” has no universal answer. A score is meaningful relative to your question set, competitive context, and whether late-stage evaluation questions are included.

How to track brand mentions on each buyer question

Manual spot-checks answer a narrow version of “how to check if ChatGPT mentions my brand.” One model, one phrasing, one moment. That workflow is slow, non-repeatable, and sensitive to phrasing. The same question in Gemini or Perplexity may name a different competitive set.

Instrumented tracking asks: on our configured buyer questions, which models mention or recommend us across scheduled snapshots? You define a set that mirrors how prospects evaluate vendors. ShareOfAsk runs those prompts across OpenAI, Anthropic, Google, xAI, and Perplexity using neutral user-role prompts without system prompt injection. Question Presence records status per question and per model.

Web mention tools track pages, not model answers. Rank trackers measure search positions, not assistant responses. An AI visibility instrument is configured around buyer questions and multi-model snapshots. Tracking presence tells you what models returned. It does not guarantee that more snapshots will make a model recommend you.

What buyers actually ask AI about products in your category

“What is the most common question people ask AI?” is the wrong frame for measurement design. Buyers in your category ask stage-specific questions: what to evaluate, who belongs on a shortlist, how options compare under constraints, what breaks at scale, and what procurement needs before a signature.

Category entry sounds like “what should I evaluate for [job to be done]?” Shortlisting sounds like “best [category] for [segment].” Comparison, risk, fit, and procurement-adjacent questions follow the same pattern — real constraints, not abstract curiosity.

The instrument only measures the questions you configure. Question Presence maps coverage for that set, not everything anyone might ask about your market. Sales calls, win/loss notes, and support themes beat generic listicles as inputs. “What questions should I ask a customer about a product?” points at the same source: questions prospects raise in evaluation are the questions your snapshot should include.

Custom prompts shape what models return; neutrality limits are in “What we measure and what we do not claim” below. See how we measure and what we do not claim for neutral measurement principles.

What presence gaps tell you (and what they don't promise)

A blank cell on Question Presence names a buyer question where models did not mention your brand — or named a competitor instead. That signal marks which buyer questions to read next in Mentions and which cells to track over time — not which page or campaign will earn a mention. It is not a causal playbook for “how to be visible on ChatGPT.”

How to get or do brand mentions usually means influence playbooks; presence measurement shows whether mentions exist on configured questions — not whether a content or PR action will create them.

A gap shows where the current answer set omits you. It does not prove which page or campaign would flip the outcome.

When competitors are configured, absence paired with Competitor Replacement Risk (CRR) shows who appeared instead and whether they were recommended rather than merely mentioned. CRR requires competitors in the project. ShareOfAsk also preserves Mentions — snippets from raw responses — so you can read how a model framed the category when your brand was missing. Recommendations, evidence-backed action plans from gaps, are Pro only.

What we measure and what we do not claim

Question Presence shows whether and how models mentioned your brand on configured questions, per model, across scheduled snapshots.

Question design matters. Manually written questions are not auto-validated for neutrality. Treat custom questions as experiments, not default truth.

Snapshots can be partial. Processing does not always complete every model on every run.

Presence records what models returned. It does not prove any source URL caused a mention. Sources show which domains appear in answers. They do not show that a cited page made the model name your brand.

A practical starting point for your category

List eight to twelve buyer questions by funnel stage, using language from sales calls rather than only SEO keywords. Generic “ten questions to ask” lists miss evaluation stage — span category entry, shortlist, comparison, risk, and procurement, not ten variants of “best [category].”

Run one neutral phrasing per question in two models manually and note mention, recommend, or absent.

Compare those results to your executive visibility number. If the rollup looks healthy but late-stage questions show absence, the headline score is hiding the gap.

Instrument the two questions where absence would hurt a deal first.

Frequently asked questions

What is a good AI visibility score?

No universal threshold exists. For leadership reporting, pair the rollup with a coverage map or a count of blank cells on late-stage evaluation questions. A strong average with gaps on integration or procurement questions is a warning, not a win. Per-question presence is the sanity check behind the headline number.

Which tool is commonly used for monitoring brand mentions?

Teams use manual ChatGPT checks, web mention platforms, SEO rank trackers, and AI visibility instruments — each measures something different. ShareOfAsk is built for multi-model, question-configured snapshots with raw response context. It is not social listening or a rank tracker.

How to be visible on ChatGPT?

When stakeholders treat “ChatGPT visibility” as a content campaign, open Mentions for a blank cell and read what the model said instead of your brand. Competitor names and perceived gaps in that framing inform positioning — not a guarantee that more pages will change the next snapshot. That beats guessing which blog post “fixes ChatGPT” without evidence.

What's a good question to ask about AI?

“Will AI replace my job?” or “What can ChatGPT do?” are curiosity prompts — they tell you nothing about whether models recommend your product on evaluation questions. A useful measurement prompt sounds like a buyer with a deadline: “Which [category] tools pass SOC 2 for a 50-person team under $X/month?” Name your category, segment, and constraint; vague popularity questions optimize the wrong rows on a coverage map.

How to find unlinked brand mentions?

Unlinked mentions here live inside model answers, not on your site. Per-question presence shows whether models mentioned you and preserves context in raw responses. That differs from finding unlinked web citations. See how we measure and what we do not claim.

To see how the measurement works, including what we do and do not claim, read our methodology. When you are ready to inspect your category, Get started and explore the demo.