Ask ChatGPT the same “which brand should I buy?” question a hundred times, and you’ll rarely get the same list twice. In a January 2026 study, SparkToro and Gumshoe.ai had 600 volunteers run 12 recommendation prompts through ChatGPT, Claude and Google’s AI nearly 3,000 times. For ChatGPT and Google AI, the odds of two responses returning the same brand list were less than 1 in 100. Matching the order as well was closer to 1 in 1,000.
That finding changes how anyone should read an AI visibility report. If answers shift on every run, a single screenshot of ChatGPT recommending your competitor proves very little. What matters is the pattern across many runs and many questions. That’s the job tools like Tesseract were built for. AdLift’s platform tracks how a brand shows up across ChatGPT, Gemini, Perplexity, Claude, DeepSeek, and Google’s AI Mode, and refreshes the data weekly.
Each of its core metrics answers a different question. Read in isolation, any one of them can mislead you.
Start With How Often You Show Up
Visibility percentage
This is the headline number. It tells you what share of your tracked prompts include your brand. Given how much individual answers wobble, a frequency like this is one of the steadier signals available, since it reflects repeated presence rather than one lucky (or unlucky) response.
The catch is that the number only describes the questions you chose to track. A project full of broad “what is” questions will score very differently from one built around head-to-head product comparisons. When the percentage moves, open the individual prompts before celebrating or panicking.
Mentions, platform by platform
Tesseract breaks brand mentions down by AI platform. That split often exposes something an overall score hides. A brand can come up reliably in Claude’s answers while barely registering in Gemini’s, and averaging the two papers over the gap.
When one platform lags, look at which prompts are missing you there. The gap might cluster around one product line, one audience, or questions asked late in the buying process.
Then Check Whether AI Treats You as a Source
Being named and being cited are different events. An answer can mention your company in passing without linking to anything you published, and that difference tells you a lot about whether your content is doing the work.
Tesseract reports citations as a running count of cited appearances across your tracked prompts. A rising count is worth digging into. It shows which pages earned the credit and for which questions. A low count alongside frequent mentions points to a different problem: AI systems know your name but lean on someone else’s pages to explain your category. Find out whose.
Volume alone can still hide where you’re missing. That’s why the dashboard also splits tracked prompts into cited and uncited. Picture a company that gets cited every time someone asks how its product type works, yet goes uncited whenever someone asks which provider suits their situation. The total looks healthy while the company disappears at the moment a buyer is deciding.
Group the uncited prompts and look for a pattern before rewriting anything. Ten uncited questions might trace back to one thin comparison page. Fixing that is a very different job from fixing ten separate gaps.
Google adds another wrinkle. Ahrefs compared about 730,000 query pairs and found AI Mode and AI Overviews cited the same URLs only 13.7% of the time, even when their answers reached similar conclusions. Tesseract now tracks AI Mode and AI Overviews separately, which matters if your citations look strong in one and thin in the other.
Put Every Appearance in Context
Share of voice
Showing up is one thing. Showing up more often than your rivals is another. Share of voice compares your presence in AI answers with competitors and other sources across the same tracked questions. You add competitor domains when setting up a project, and from then on the comparison runs alongside your own numbers.
This is where comfortable stories tend to fall apart. A provider might appear in nearly every answer explaining its category, yet rarely show up when the prompt asks which provider to choose. Its visibility percentage looks fine. Its share of voice at the decision stage tells a less flattering story. Review the competitor-heavy answers and the sources they cite, because those sources are usually where the gap starts.
Sentiment
An appearance count can’t tell you how you were described. Tesseract includes sentiment data for that reason. Use it as a flag that a response deserves a closer look. It shouldn’t replace reading the response.
Being listed as “a solid option for small teams” and being named in a paragraph about common complaints both count as appearances, and they mean very different things. When sentiment highlights an answer, read it in full. Check what was asked, which other brands appeared and which sources the answer leaned on, then decide whether anything actually needs fixing.
Your Prompt List Decides What the Report Can Tell You
Every number above sits on top of one choice: which questions you track. Tesseract lets teams combine their own custom prompts with suggested search intents, and pick separate keywords for monitoring Google’s AI Overviews.
Choose carefully, because the report can’t see beyond that set. Results describe performance on those prompts, not on every conversation people have with AI about your category. The most useful prompt lists mirror how real buyers talk: the problems they describe, the comparisons they request and the use cases they mention. Sales calls and support tickets are better sources for these than a keyword tool. Content built around real buyer questions is also what tends to earn a place in AI answers, which is the core of most marketing content for AI search strategies.
The SparkToro research adds one more reason to choose well. Since individual answers vary so much, a decent number of prompts tracked over several weekly refreshes tells you far more than a handful checked once.
From Dashboard to Next Move
A practical review order looks something like this. Glance at visibility percentage for direction. Use mentions and citations to see where you appear and where you’re actually credited. Bring in share of voice and sentiment to understand the company you keep in those answers. Then go prompt by prompt to the places where competitors surface and you don’t.
From there, pick one gap that maps to real revenue. Read the answers, note the sources being cited instead of you, and improve the page that should be answering that question. Watch the prompt over the next few refreshes. Visibility on its own doesn’t pay the bills, so it helps to know which tracked questions feed demand and lead generation and which only build general awareness. Tesseract can connect GA4 and Search Console for that side of the picture.
One question is worth sitting with before you open the dashboard: if AI answers barely repeat themselves, how many runs of a prompt would it take before you trusted a change enough to act on it?
Related: SocialSight AI 2026: Plans, Pricing &Tokens or Worth It?
