You want to know whether people are recommending your restaurant at dinner parties. You cannot attend the dinner parties. You can count bookings, ask new customers how they heard about you, and occasionally get invited to a party and listen.
That is the situation exactly. There is no dashboard for a conversation you are not in. There are three imperfect instruments, and using all three beats trusting any one.
Instrument one — your own prompt set (the most valuable, and free)
Build a fixed list of 20 to 40 questions your buyers actually ask, run them on a schedule, and record what happens. This is a research panel, not a ranking report, and treated that way it is the most honest signal available.
Not keywords. Full questions, in natural language, the way someone types into a chat box. "Who can handle DG cargo clearance at Port Klang?" not "DG cargo clearance Port Klang".
Informational ("what is…"), comparative ("X vs Y", "alternatives to…"), instructional ("how do I…"), brand ("what is [you]", "is [you] any good"), and transactional ("best [category] in [place]"). Tag each by funnel stage so you can tell an awareness gap from a shortlist gap.
Logged out, no history, incognito. Otherwise the engine is answering for you, not for a stranger, and you will flatter yourself.
Were you mentioned (yes/no). Was the mention accurate (yes/no). Who else was mentioned. Which domains were cited. Nothing else. Four columns in a spreadsheet.
The absolute number is meaningless. The trend, and the competitor comparison, is the whole point.
Two derived numbers are worth tracking: your presence rate (the share of prompts where you appear at all) and your share of voice (your mentions as a proportion of all brand mentions across the set). Both are crude. Both move in the right direction when the work is working, which is all you need from a metric.
Instrument two — Search Console and server logs
Google launched Search Generative AI performance reports in Search Console on 3 June 2026, giving a dedicated view of impressions inside AI Overviews, AI Mode and generative features in Discover. TIER A Know its limits before you rely on it:
- Impressions, pages, countries, devices and dates only — no clicks, no CTR, no query data at launch.
- Data starts 18 May 2026. There is no history before that.
- It rolled out to a subset of site owners first, beginning in the UK, expanding since.
Alongside it, your raw server logs answer a question no dashboard will: are OAI-SearchBot, PerplexityBot and ClaudeBot actually fetching your pages, which ones, and how often? This is the only direct evidence that the machines are reading you, and it is free.
Instrument three — referral traffic, read correctly
Segment sessions from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Then apply three corrections before you believe anything:
- It undercounts, structurally. Many AI referrals arrive with no referrer header and land in "direct". Your true AI traffic is higher than the number you see, by an amount you cannot determine.
- Engine splits are unreliable. Because of that stripping, the apparent breakdown between engines is partly an artefact of which ones happen to pass a referrer. Do not make decisions on it.
- Judge it on quality, not volume. Conversion rate and value per session, compared against organic. Volume will look disappointing for a long time and that is not the story.
Any tool claiming to show your "ranking" inside an AI model, your "AI algorithm score", or data drawn from a platform's internals. Google states directly that no third-party tool has access to its internal ranking or AI systems. TIER A Every legitimate visibility tool works the same way you would manually: it runs prompts, at scale, and counts what comes back. That is genuinely useful — it is sampling, not telemetry, and a vendor who blurs the difference is telling you what they are.
Should you buy a tool?
The category — Profound, Peec, Otterly, Semrush's AI toolkit, Ahrefs Brand Radar and others — automates instrument one across more engines and more prompts than you would run by hand, and adds competitor tracking. That is real value at scale.
My honest advice for a business under, say, fifty people: run the spreadsheet by hand for three months first. You will learn more from reading forty answers yourself than from any dashboard, you will discover that your assumed competitor set is wrong, and you will know exactly what you need before you pay for it. Then buy a tool if the manual version has become the bottleneck.
Create a spreadsheet with 20 real customer questions and four columns. Fill it in this month. Fill it in again next month. That is a measurement programme.