1. Analyze a brand. Type the brand name or its website — Hulu or hulu.com — and press Analyze. If the name is ambiguous, pick the right company in the popup.
2. Choose the setup. Step 1: the language your customers ask in (Auto-detect reads it from your site). Step 2: your target market — the country whose web results the AI should see, or Global.
3. Review the proposed study. We suggest category questions (never naming any brand) and real competitors from the live web. Remove what doesn't fit, add your own questions or rivals.
4. Confirm & analyze. The set freezes (that's what makes trends comparable) and the run starts in the cloud — follow it in the tray, or leave; it finishes on its own in a few minutes.

1. Start at the top. Visibility = % of AI answers that mention you; the shaded band is how sure we are. Category ranking tells you who owns the conversation. From the second run, Change vs last run says if you moved for real (bands apart) or within noise (bands overlap).
2. Find where you're missing. The Mentions-per-question matrix shows exactly which questions you win (✓) and which you lose (—). Click view answers to read each AI answer as a chat and see who got recommended instead of you.
3. Understand the why. Perception shows how the AI talks about each brand, with verbatim quotes. Sources shows which sites informed the answers — if yours isn't there, that's the root cause.
4. Act on it. The GEO action plan and Optimize turn the gaps into content moves. Then turn on Monitoring so the same study re-runs on schedule, and annotate launches to explain future moves in the trend.

When people ask AI assistants things like "what's the best platform for X?", some brands get recommended and others don't exist. GenSig asks the AI real category questions — without ever naming your brand — and measures how often you appear versus your competitors.

Visibility is the share of AI answers that mention your brand. That's the number everything else builds on.

• Type your brand name or website — Hulu or hulu.com both work. If a name matches several companies, we ask which one you mean.
• Pick the questions language (or let us auto-detect it from your site) and your target market — any country, or Global.
• Review the proposed questions and competitors, edit anything, and press Confirm & analyze. That configuration gets frozen so every future run is comparable.

The analysis runs in the cloud — you can navigate away or close the tab, and the tray at the bottom right shows progress.

Every question is asked on the live, web-grounded surfaces you choose for that analysis — ChatGPT with real web search, Gemini with Google Search grounding, Perplexity with its own live search — the channels that mirror what real users see. Each surface runs the full question set in your target market's context (some more than once per question), and each one counts as a run from your allowance.

We validated this in our own published research: asking a bare model (no web access) measures the model's memory, which lags the market — it misses recent brands and inflates stale ones. Live surfaces don't.

AI answers vary between runs, so a visibility number from a handful of answers carries uncertainty. The shaded band shows where the true rate most likely falls (a 95% confidence interval): the dot is what we measured, the band is the plausible range.

• Narrow band — solid reading (high precision).
• Wide band — few answers; add questions to tighten it.

The card stacks your previous band over the current one. If the two bands don't overlap, the change is statistically distinguishable in this study — larger than the sampling variation we observed, so the delta shows in green (up) or red (down). If they overlap, the change is within the noise: not distinguishable yet; more runs will tell. It's a signal about these prompts on these surfaces, not proof the whole market moved or that a specific action caused it.

Because they genuinely do — each assistant searches differently and favors different sources. There is no single "what the AI says": there's what ChatGPT says, what Gemini says, what Perplexity says. The chips at the top of the report show which surfaces that run measured and how each one scored; expand any question's answers to read them like a chat.

Two runs are only comparable when they measured the same surfaces. That's why a run on different surfaces shows "No comparison" instead of a change — a delta between different surfaces measures the surface, not your brand.

Your questions and competitor list freeze when you confirm them, and every run uses the same set — that's what makes the trend comparable over time. Editing them bumps the set version (shown on each report), so you always know which runs can be compared apples-to-apples.

Competitor suggestions come from the live web, localized to your market — not from the model's memory. You can add or remove them before confirming.

The language controls how the study's questions are written — pick the one your customers use. The target market localizes the actual web searches the AI performs: results in Argentina and Spain differ a lot, even in the same language. Any country works, or choose Global for a worldwide read.

Where the AI gets informed lists the domains the AI actually cited while answering — with real example URLs on the Sources page. If your own site isn't among them, the AI is learning about your category from third parties.

The GEO action plan and the generators in Optimize turn that into concrete content moves: where to appear and what to publish so the AI starts citing you.

Turn on Monitoring to re-run the exact same analysis on a schedule (weekly, bi-weekly or monthly) — the trend chart grows by itself. Each scheduled run counts against your allowance like a manual one.

Use Annotations to mark launches, campaigns or PR on the timeline, so visibility moves have context later.

• Pro: 3 brands · 40 analyses per month (weekly billing: 10 per week, resets Monday UTC).
• Agency: 15 brands · 150 per month (weekly: 35) · team of 10.

Monthly and annual plans reset on the 1st (UTC). Scheduled monitor runs count against the allowance too. Usage shows exactly where you stand.

Out of allowance mid-week? Buy a one-time pack — 10, 25 or 60 extra runs. They never expire and are consumed automatically only when your plan allowance runs out, newest analysis first. Your banked balance shows in Usage. No subscription changes involved.

• Owner — everything, including billing and member management.
• Editor — can run analyses and edit the brands they're given access to.
• Viewer — read-only: sees reports, trends and annotations, can't change or run anything.

Roles are per brand, so an agency can give each client's team access to just their own dashboards.

Write to [email protected] — refunds, downgrades, higher limits and anything this page didn't answer.