Mentioned is nice. Recommended is better.
For every mention, a language model with a fixed rubric decides whether the tone is positive, neutral or negative, with the theme and the exact sentence as proof. You see the breakdown per week, per AI model and per business.
ChatGPT
Perplexity
Gemini
ClaudeA mention means little until you know how you're described.
Visibility tells you whether an AI model mentions your business. But someone asking ChatGPT which installer to hire doesn't read a list of names. They read a verdict. “Great reviews, but long wait times” is still a mention, and that one sentence can cost you the job. Sentiment shows you what's behind every mention.
AI models build that verdict from what's already been written about you: reviews, comparison sites, forums, news. One critical source that gets cited often can color the picture across all four AI models. Without tracking it, you only notice when the leads slow down.
GeoReady rates every mention on its own, groups the claims by theme, and attaches the exact sentence from the answer to every verdict. You don't just learn whether the picture is right. You see where it comes from and what to do about it.
“Getting mentioned is step one. Getting recommended is where customers decide.
Every mention gets a verdict, a theme and proof
After every measurement, a language model with a fixed rubric reads each sentence that mentions your business. It decides whether the tone is positive, neutral or negative, and which theme the sentence is about: price, quality, service, delivery, reliability, expertise, availability or sustainability. Anything that doesn't fit lands under “general.”
The verdict is never a black box. Every claim shows the exact sentence from the AI answer, the AI model that gave it and the date. One click takes you to the full answer and the prompt that produced it, so you can always check the call yourself.
Every classification has a confidence level. At the bottom of the page you see the average confidence for the period and how many verdicts were corrected by hand. You know how solid the numbers are before you take them into a meeting.
ChatGPTPerplexityGeminiClaude- Read a one-line summary up top: what share of your mentions is positive, what share is negative, and what the negative ones are mostly about.
- See the exact sentence, AI model and date for every claim.
- Jump from any claim to the full answer (Chat →) or the prompt behind it.
- Check the average classification confidence and the number of manual corrections.
Know what the praise and the criticism are about
A positive percentage is a start, but you can't act on it until you know what it's about. That's why claims sit in three columns (positive, neutral, negative), grouped by theme, with a count per theme and the first sentence as a preview.
Expand a theme to see every claim underneath it, up to forty per theme. Filter by period or AI model to see the rest. The biggest negative theme opens by default, because that's usually where the fastest wins are.
Themes turn a feeling into a plan. “AI is negative about us” goes nowhere. “Gemini mentions our maintenance wait time three times” turns into a page, a review reply or a conversation with your scheduling team.
- See positive, neutral and negative claims side by side, grouped by theme.
- Get a count and an example sentence per theme without clicking anything.
- Expand a theme for every claim, with AI model, date and links to the chat and prompt.
- Narrow the list with the period and AI model filters.
Each AI model has its own picture of your business
ChatGPT, Perplexity, Gemini and Claude don't read the same sources or summarize the same way. One AI model can be mostly positive about you while another is critical. The page shows the positive, neutral and negative split for each AI model, with a sentiment score from 0 to 100.
The AI model with the highest share of negative mentions (with at least three mentions) is called out by name at the top. It's often the model citing one specific critical source, and the chats show you which one.
A stacked chart shows the split week by week. You can tell whether a shift is a one-off or a trend, and whether a fix starts showing up in the tone a few weeks later.
ChatGPT · Oct 8 3:10 AM · run 3/3mentioned #2 · positiveA few installers stand out: Summit Climate Co. for large projects, Northfield Heating & Air from Manassas for “clean workmanship and an honest quote” [reddit.com] [northfieldheatingair.com], and Redwood Home Services for hybrid systems. If you want a tax credit, check the federal 25C requirements…
- Compare the sentiment score (0–100) for each AI model at a glance.
- See which AI model is most critical of you.
- Track the positive, neutral and negative split per week in a stacked chart.
- Filter the whole page to one AI model to read only its claims.
- Set an alert under Alerts for when your sentiment score drops below a threshold.
See what AI says about your competitors
Sentiment isn't only tracked for your business. It's tracked for every business that shows up in the answers. The per-business block puts your split next to each competitor you follow, on the same 0–100 score.
Click a competitor and the claims below switch to what AI says about that business, by theme and with proof. You see what a competitor gets praised for, and where they take criticism you can use.
That's useful input for your positioning. If AI praises a competitor for fast service and you for craftsmanship, you know which story to strengthen and which gap you can fill.
- See your sentiment score next to every competitor you follow.
- Click a competitor to read every claim about that business, by theme.
- Switch back to the claims about your own business in one click.
- Compare what competitors get praised for with the themes you win on.
Correct a verdict, then fix recurring criticism
A language model can misread a sentence, like sarcasm or a line that's really about another business. So you can change the verdict on any claim to positive, neutral or negative. The correction flows straight into the numbers and is labeled “manual,” so everyone can see a person reviewed it.
When a negative theme keeps coming back, GeoReady suggests an action after the measurement: address the negative theme, with how often AI brings it up and a step-by-step plan. It works the other way too. If AI keeps praising you for something, you get an action to back it up with proof on your homepage and in your reviews, so it shows up more often.
Thirty days after you mark that action done, GeoReady measures the effect on the related prompts. You see whether the fix actually made it into the answers.
- Change the verdict on any claim from the dropdown. The numbers update right away.
- Spot corrected claims by their “manual” label.
- Find the suggested recovery action for a recurring negative theme under Actions.
- Turn positive themes into action: AI praises you for something, so prove it on your site.
How Sentiment Analysis works.
How Northfield Heating & Air fixed a complaint that wouldn't go away
Northfield Heating & Air installs heat pumps across Northern Virginia and gets mentioned a lot. Still, its AI-driven quote requests lagged behind Summit Climate Co.
- 1Spot the pattern
The sentiment page showed 62% positive and 18% negative. Most of the negative was about “availability,” and Gemini was the most critical: seven answers said Northfield was “hard to reach for maintenance.”
- 2Find the source
Chat → showed that Gemini kept citing a review page on heatpumpcompass.com with three old complaints about phone availability from last year.
- 3Fix it
Northfield replied to the reviews, published its new maintenance scheduling with online booking on its own page, and marked the suggested action “Address the negative theme ‘availability’” as done. The team also corrected one claim that was really about Redwood Home Services to neutral.
- 4Measure the effect
Thirty days later, the availability theme in the negative column was down from seven claims to two, and the sentiment score on Gemini rose from 54 to 68.
Every number traces back to the exact prompts.
Questions about Sentiment Analysis.
Another question? Email info@geoready.nl.How reliable is the sentiment analysis?+
Every classification has a confidence level. Borderline cases are flagged as such, and you can correct them.
Can I see which source drives a negative take?+
Yes. The chat shows the cited sources, and the action shows which source to correct.