Sentiment Analysis
How AI engines describe and rate your brand and the competitors you track, with the quotes and pages behind every finding.
When a buyer asks an engine whether you're any good, the engine researches you: reviews, forums, articles. Sentiment Analysis shows what it found and said, about you and every competitor you track, in the engine's own words and with the pages behind each finding. The ratings are ChatGPT's own.
What it shows
The page is a set of cards you can rearrange:
- Categories Leaderboard: your place in up to four categories, and who leads each.
- Top Strengths and Top Concerns: the strongest positive and negative findings about you.
- Position vs Competitors: every brand's rating in every category, side by side.
- Most Praised and Most Criticized: how many positive or negative findings fall in each category.
- Perception Drivers: the sites behind the positive and negative findings about you.
- Sites That Cover Competitors, Not You: sites that write about your competitors and leave you out.
The strengths, concerns, praised and criticized cards can be added for any tracked competitor, with its name in the title, such as "Top Concerns, Acme".
Every row opens its findings: how each engine said it, and the pages behind it. Methodology, in the toolbar, sums up how the page is built. Until the first check finishes, the page says so.
How it is counted
Sentiment checks
A sentiment check asks each engine the same six questions about your brand and about every competitor you track. Web search is on, so the answers rest on the reviews, forums and articles the engines read. The six questions cover:
- Overall perception: how the brand is seen and known.
- Strengths: what customers and reviewers praise.
- Weaknesses: what they complain about.
- Customer experience and value: price, buying, use, support and delivery.
- Best and worst fit: who it suits and who is better served elsewhere.
- Competitor comparison: where it is stronger or weaker, and for whom.
The exact wording stays private. No question asks for a sentiment or a score, only what customers, reviewers and the press report, with sources.
A full check asks the six questions about once a month, on your brand's engines apart from AI Overview.
Findings
Every clear claim in the answers becomes a finding. A finding is kept only with a verbatim quote from the engine that said it.
Findings that say the same thing about a brand are one idea, counted once: "answers within a day" and "fast, helpful support" become one row with both quotes. Over a year, an idea later checks find again still counts once, with how many checks found it.
Each finding has a tone, positive, negative or mixed, and one category. Tone is read from the asked brand's side: in a comparison, a competitor doing better is negative for the brand.
Counted, not scored
Findings are counted, never turned into a score: the page tells you how many positive and negative ideas the engines raised, and which were raised most widely. The ratings are separate. ChatGPT gives them when asked on its own, they are not worked out from the findings, and the two need not agree.
Categories
Findings and ratings use ten universal categories, the same for every brand: quality and performance; features and capabilities; reliability and durability; ease of use; price and value; customer service and support; purchasing, delivery and access; trust, safety and responsibility; customer fit and use cases; brand reputation and positioning.
Your brand also has industry categories, specific to what you sell. They are built once, from the findings of your brand's first check, and then stay fixed so readings compare over time.
Ratings
ChatGPT rates each brand in each category from 1 to 10, from what it finds online. It is asked every week. A cell shows the average of every rating in the period you pick.
| Rating | Band | Pill color |
|---|---|---|
| 9 to 10 | Very positive | Green |
| 8 | Positive | Light green |
| 7 | Positive | Yellow |
| 5 to 6 | Neutral or mixed | Orange |
| 1 to 4 | Negative | Red |
The pill's color comes from the average rounded to a whole number: 7.4 reads as 7, positive, in yellow.
A dash means ChatGPT did not rate that brand in that category. Hover it for the reason: usually not enough independent information online, and now and then a call that failed. A dash is not measured, not a low rating.
Ratings from before 9 October 2026 were on a 1 to 5 scale. They are never averaged with later ones, and a period holding only them shows a 1 to 5 legend.
Biased
Biased marks a brand whose findings or ratings rest mostly on its own website. What a company says about itself is not what customers or reviewers think. The badge has one meaning and three thresholds:
| Where | Biased when |
|---|---|
| Findings, on this page | at least half of a brand's ideas of one tone rest only on its own pages |
| Ratings, on this page | at least 80% of the pages behind a brand's ratings are its own site |
| Citations, on Overview and Visibility | at least 80% of the pages engines cite about a brand are its own site |
Biased ideas still count. Their pill reads "n own pages" in place of a source count. Your own pages are those on your domain and on the other websites you list in General.
What the period does
- Ratings average every check inside the period: the newest for the last run, the week's for 7 days, about four for 30 days, every one for the year.
- Findings come from full checks, about one a month. The last run, 7 days and 30 days read the newest full check up to the period's end. The year reads every full check in it and merges the same idea across them, so a finding can read "found by 3 of 11 checks".
If no check falls in the period you chose, the page says so and links to the latest.
How to read it
About this sentiment check
The info dot beside the line under the title says which checks the page reads: your brand, the market, when its checks were asked, how many brands and engines took part, and how many findings came back. It also says how many findings rest only on the brands' own pages, and how many questions got no answer.
Categories Leaderboard
For each category you pick, up to four of the universal ones, the card shows your place among the brands ChatGPT rated there, such as "#2 of 7", and the leader. Brands are ranked by their average rating over the period. Brands with the same average share a place, and your place then reads "tied". If you were not rated in a category, the card says "not rated" rather than giving you a place.
Click a category for the full leaderboard: every category and every rated brand in rank order, with the average each rank rests on. The card needs at least one tracked competitor.
Top Strengths or Concerns
Top Strengths and Top Concerns list the strongest positive and negative findings about a brand, four at a time. A finding ranks higher when more checks found it, then when more engines said it, then when it came up in more of the six questions, then when more pages back it.
Each row shows the finding's tone and how many sources back it. Open a row for each engine's own words and the pages behind them. The expand button opens the full list.
A Biased badge beside the title means at least half of these ideas rest only on the brand's own pages; hover it for the count.
Position vs Competitors
Each cell is ChatGPT's 1 to 10 rating of one brand in one category, averaged over the period and colored by band (see Ratings). Your column is highlighted.
- Hover a cell for the latest rating and ChatGPT's reason.
- Click a cell to open the findings behind it. They come from full checks and need not agree with the rating.
- Click a brand's name to sort the categories by its ratings, highest first, then lowest first.
- Switch between universal and industry categories at the top of the card. Industry categories show wherever you were rated, and competitors they do not fit stay empty.
- A dash means not measured: not enough independent information online, or a failed call.
- A Biased badge under a brand's name means at least 80% of the pages behind its ratings are its own site.
"All competitors and categories" opens the table at full size, with filters for competitors and categories.
Most Praised or Criticized
Most Praised and Most Criticized count a brand's positive or negative findings per category, most first, leaving out categories with none. Each finding is one idea, counted once however many engines said it. The expand button opens every category with its findings. Biased works as on Top Strengths or Concerns.
Perception Drivers
The sites the engines cited for the positive and negative findings about you, busiest first. These are the sites shaping what an engine says about you. A site counts once per finding.
- Findings: the bar splits the site's findings into positive and negative. Hover it for the counts.
- Engines: how many of the check's engines cited the site, such as 2/3.
- Type: what kind of site it is, such as a review or news site.
Your own pages are never drivers; each finding lists them apart. The link under the table opens every site, with filters.
Sites That Cover Competitors, Not You
Orvio reads every page the engines cited for findings about your competitors. A site is listed when one of its pages names a tracked competitor and does not name you.
- Names: the competitors the site's pages name.
- Pages read: how many of the site's cited pages Orvio could read, out of how many were cited.
- Engines: the engines that cited it.
A row may add "names you on N other cited pages" when another page on the site does. Open a row for the qualifying pages.
A page Orvio could not read counts neither way. Your site and your competitors' own sites are never listed. While pages are still being read, a spinner shows beside the title and the list may grow.