GenSig measures how visible your brand is when AI assistants answer your customers' questions, and gives you the evidence, the sources and the plan to improve it.
When someone asks ChatGPT "what's the best specialty coffee subscription?", the answer either names your brand or it doesn't. GenSig asks the questions your category gets asked to AI assistants with their live web search on (we call each one a surface): ChatGPT, Gemini, Perplexity, Claude, Grok and Mistral, through their official APIs. The questions are localized to your target market: a frozen, versioned set of brand-neutral questions (8 by default, up to 30 per brand). Surfaces without an API are covered by human measurement. It measures:
I got a report from my agency. Who is GenSig?
GenSig is the tool your agency uses to measure how ChatGPT and other AI assistants name your brand. Every number comes from AI answers to your customers' questions and shows its margin of error, so you can see how sure each result is. Your agency prepared the report and can walk you through any number.
Your questions are generated once, reviewed by you, then frozen and versioned. Every future analysis asks exactly the same questions. Otherwise the trend would compare different questions. Editing questions or competitors bumps the version, and the chart warns you when a period mixes versions.
Every question is asked on channels that mirror what users see: ChatGPT with live web search, Gemini with Google Search grounding, Perplexity, Claude with web search, Grok with live search and Mistral with its web connector, all through official APIs and geolocalized to your target market. We published the experiment behind this design: 2,500 measured answers showing that bare LLM APIs measure the model's memory (they miss recent brands and inflate stale ones), while each search-grounded channel closely mirrors its own consumer product.
You pick the surfaces for each analysis, and each surface produces its own report and its own trend line. One analysis asks every frozen question once on that surface. Each AI answers differently, so GenSig reports each one separately: comparisons and changes are computed only between analyses of the same surface and question-set version. AI answers vary from one analysis to the next, so that variation is shown as confidence bands, and trends over repeated analyses matter more than any single point. Claude is the most expensive surface to measure, so an analysis on Claude uses 3 analyses of your allowance; every other surface uses 1.
A single visibility number hides where you actually lose. Each question in your set is classified by intent: discovery ("what tools exist for…"), comparison ("X vs Y", best-of lists), purchase ("which should I use for…"), problem ("how do I solve…") and reputation ("is this trustworthy"). Visibility is then broken down by intent. Being strong in comparison questions but invisible in purchase questions is a very different problem from the reverse, and it calls for different work.
Every visibility number ships with a shaded band: the range where the true rate most likely falls (a 95% Wilson interval), plus a plain-language precision label. The change against the previous analysis shows both bands. GenSig marks a change only when it is bigger than the margin of error (a two-proportion test at 5%). Otherwise it says No detectable change yet instead of celebrating noise. For example, with 12 questions, going from 7 of 12 answers to 9 of 12 is not a detectable change; going from 2 of 12 to 9 of 12 is.
Each archived analysis is compared with the previous comparable analysis: same brand, same question-set version, same surface. If no comparable analysis exists, the report says "No comparison" and explains why. It never compares a Gemini analysis with a Perplexity one. That is the number behind the +/- badges.
Every recommendation in the action plan is derived from counts measured in that analysis, not from a model's opinion: which domains informed which questions, in how many of those answers your brand was weak or absent, and which competitors those same answers name instead. The numbers are computed from the archived answers; the language model only phrases the next step, and only from that evidence. It is also told not to give generic advice that would apply to any website.
What GenSig deliberately does not claim: causal verdicts. Visibility moves for many reasons, not just your actions: provider updates, source churn, your competitors' moves. That is why actions are tracked as Annotations on the timeline, so you see the correlation with your own eyes instead of getting a made-up "your action won you 3 more answers".
Of the 33 surfaces in the catalog, only six expose their answer engine through an official API. The rest, such as Google AI Overviews, AI Mode, Meta AI on WhatsApp, Copilot, Siri, DeepSeek, Luzia and Zapia, can only be seen inside the app. GenSig measures them with people:
Human Verified is in early access. Write us for a slot.
| View | What it shows |
|---|---|
| Dashboard | The active brand's latest numbers, the visibility trend with annotation markers, and the full analysis history (paginated). Start new analyses from here. |
| Prompts | The frozen question set: version, language, per-question source, draft editing with explicit Save. Changing it bumps the version. |
| Competitors | Side-by-side visibility, sentiment and position for every tracked brand, Share of Voice over time (you vs the top competitors, filterable by surface), plus an on-demand autopsy of any competitor's site examined with the same lens as yours. |
| Perception | What the AI actually says: verbatim quotes grouped into praise, criticism and neutral mentions, per brand. |
| Sources | The domains AI cites when answering your category questions, classified (editorial, UGC, competitor, you…) with example URLs, sources over time (new, lost, rising, falling per analysis), plus source opportunities: sites that inform several of your questions while the answers rarely mention you. |
| Action plan | Prioritized tasks with the measured evidence behind each one: crawler blocks, citation gaps, position gaps, perception gaps, source and intent gaps, and surface gaps (named by one AI, missing on another). Each comes with an evidence level (verified / correlational / unproven), the questions affected, who wins them instead, and the concrete next step. Includes a zero-cost check of whether AI crawlers can read your site (robots.txt and llms.txt). Built from API analyses: human sessions carry no cited sources. |
| Content | Draft a citable article from a measured content gap in one click, with a live citability score, structure checks, a saved library, and export to Markdown. Pro includes 100 drafts a month and Agency 400. |
| Optimize | Generated JSON-LD structured data and meta descriptions built from your frozen questions and your site's content. Copy, paste, ship. |
| Human verified | Turn weekly human measurement on per brand, pick up to 8 human surfaces, and follow live progress, with the evidence (screenshot, verbatim, session link) attached to every session. |
| Annotations | Every GEO action you take, pinned to the timeline and overlaid on the dashboard chart as colored markers. |
| Usage | What your plan includes, what this period has used, what monitoring holds ahead of time, the extra analyses left, and the date the allowance restarts. |
| Activity | The workspace audit log: who changed what and when, including restorable configuration snapshots. |
| User management | Invite teammates with granular per-brand permissions. |
Pro and Agency plans can schedule automatic analyses (weekly, biweekly, monthly) per brand. Monitors work in the cloud, so your computer does not need to be on. The first analysis starts within about 4 hours of activating a monitor; results appear in the dashboard and you get a live notification when an analysis starts and finishes.
Each monitor has its own surface selection. By default a monitor measures Gemini only, so connecting new surfaces never silently multiplies your scheduled cost. You choose exactly what is measured on schedule, and the monitor card shows how many analyses each cycle uses. Each surface measured counts as one analysis of your allowance, except Claude, which counts as 3.
https://app.gensig.app/mcp as a remote connector in Claude, sign in with your GenSig account, and your assistant can read your brands, reports, visibility history, questions, competitors, content opportunities and action plans directly. The connector is read-only.Teams are available on the Agency plan (up to 10 members). Viewers can be scoped to specific brands, which is ideal for giving each client read-only access to their own numbers.
| Role | Can do |
|---|---|
| Owner | Everything, including deleting brands and analyses, restoring configurations and managing members. |
| Editor | Start analyses and edit configuration on all brands or a specific list; optionally add new brands. Editable brands are always visible. |
| Viewer | Read-only access to all brands or a specific list. |
GenSig explains why competitors are winning, which sources influence the answers, and what to publish, then turns that into a prioritized, implementation-ready action plan. Every recommendation is tied to measured evidence from AI answers (no black-box scores). GenSig never scrapes: it uses official APIs for the six connected surfaces and people with screenshot evidence for the 27 surfaces without one. Every human measurement carries a chain of custody you can show a client.
AI visibility is how often an AI assistant names your brand when someone asks it a buying question in your category, for example “what's the best project management tool?”. Unlike search rankings, the assistant returns a short list of brands, not ten links, so being named (or not) directly shapes which brands a buyer considers.
Generative engine optimization (GEO) is the practice of getting your brand named by AI assistants like ChatGPT and Gemini when they answer questions in your category. It works differently from SEO: instead of ranking a page, you influence the sources and signals the assistant draws on when it composes an answer.
You ask ChatGPT the category questions buyers ask, on a schedule, with web search on and in the target market. Then you count how often the brand appears versus its competitors, with confidence intervals. GenSig asks each question on every surface you select (ChatGPT, Gemini, Perplexity, Claude, Grok, Mistral), each through its official API with live search. Each AI answers differently, so GenSig reports each one separately.
The right tool measures the live, web-grounded surfaces users see (not a bare model API), reports uncertainty honestly with confidence intervals, and shows the evidence behind every number. GenSig does this, and publishes its methodology and the raw data of its research so the results can be reproduced. Evaluate any tool on which surfaces it measures, whether it shows the underlying answers, and whether it can prove its numbers.
SEO measures where your page ranks in a list of links; AI visibility measures whether your brand is named inside the answer itself. An assistant names a handful of brands and there is no page two, so the goal shifts from ranking a URL to being one of the few brands the AI names, and to appearing in the third-party sources it cites.
Because a bare model API answers from memory, not from what users actually see. In our published study the bare API correlated only ρ=0.23 with real ChatGPT web sessions: it missed recent brands and inflated stale ones. GenSig always measures the search-augmented surfaces, which mirror the consumer product.
A 33-surface catalog. Six are measured through official APIs with live web search (ChatGPT, Gemini, Perplexity, Claude, Grok and Mistral), each reported separately. The 27 surfaces without an answer-engine API, such as Google AI Overviews, AI Mode, Meta AI on WhatsApp, Copilot, Siri and DeepSeek, are measured by people in the apps, with screenshot evidence. GenSig deliberately does not measure bare LLM APIs and never scrapes.
Every report includes a prioritized action plan where the evidence is computed from the analysis: which sources informed which questions, where your brand is weak, and which competitors win those answers. The suggested next step is written only from that evidence, and the model is told not to give generic advice.
Weekly to monthly for most brands. AI answers shift as providers update models and as sources change, so a single measurement is a snapshot; trends over repeated analyses are far more meaningful. GenSig can repeat the same frozen study automatically on a schedule.
From the counts of the analysis itself: the questions where your brand is weak or absent, the domains that informed those answers, and the competitors those answers name. Each task shows that evidence next to the suggested action, so you can check the reasoning instead of trusting it.
AI surfaces are probabilistic and providers update their models, so answers vary between analyses. That is why GenSig reports confidence bands, looks at trends over single points, and marks a change only when it is bigger than the margin of error (a two-proportion test at 5%).
GenSig is $49/month for Pro (3 brands, 40 analyses) and $149/month for Agency (15 brands, 150 analyses, team of 10); one-time packs of extra analyses are available and never expire. Pro and Agency include the full live study, sources, action plan and monitoring.
Yes. Deleting a brand permanently removes its configuration, archived reports and monitors. Deleting your whole account (data, subscription and sign-in) is self-service under Settings → Delete account and takes effect immediately.
No. Every scheduled analysis counts within your plan's allowance exactly like a manual one: one analysis per surface, or 3 for Claude.