Key takeaways
- ChatGPT does not have a results page, so "rank" is replaced by four metrics: Visibility Score, Position, Share of Voice and Sentiment. Each answers a different question.
- Visibility Score tells you whether you are named. Share of Voice tells you whether you are named instead of competitors. Position tells you how early. Sentiment tells you how well.
- There is no universal formula. Some tools define visibility as a simple mention rate, others as a weighted 0-100 composite, so compare numbers only inside one tool and one prompt set.
- One ChatGPT answer carries almost no signal. Sample many prompts, repeat them, segment by platform and intent, and read trends rather than snapshots.
- Read the metrics together. A coverage problem, a prominence problem and a reputation problem look different, and each needs a different fix.
Why classic rank tracking breaks in ChatGPT
In Google you track a keyword and get a number between 1 and 100. In ChatGPT there is one answer, generated on the fly, and it changes between runs. Either your brand is named or it is not. Either your domain is cited or it is not. As one explainer from LLM Pulse puts it, the measurement collapses from a rank distribution across ten results to presence or absence in a single synthesised reply. That makes the math simpler and the variance much higher.
The answers are also crowded. According to Promptwatch Data for Sep 28 to Oct 4, 2026, 47.6% of ChatGPT responses with citations cite 10 or more unique domains, and 66.1% cite at least 8. Only 6% cite three or fewer. With that many sources in play, a binary "cited or not" tells you very little. Where you appear in the response, and how you are described, carries more information.
That is where the four metrics come in.
Visibility Score: are you named at all?
Visibility Score is the baseline. In its simplest form it is the percentage of tracked prompts where your brand shows up. Track 200 prompts, appear in 60, and your score is 30%. Cognizo calls it the AI equivalent of impressions, and that is a fair mental model.
Tools do not agree on the formula, though, and this causes most of the confusion:
- Simple mention rate: brand mentions divided by total relevant responses. OptimizeGEO describes a 0-100 composite built on this, plus prompt coverage, prominence, sentiment and cross-platform consistency.
- Weighted composite: Promptwatch scores each response on placement, share of attention and depth, then averages them. Responses that do not mention your brand count as zero.
- Mention plus recommendation: Visiblie separates being mentioned from being actively recommended, and a score that rises while recommendations stay flat means you gain mentions without endorsement.
There is no universal "good" score. A 30% score is strong if the category leader sits at 15% and weak if they sit at 60%. A brand that holds 30 to 35% across repeated runs has a stable position. One that swings between 15 and 45% has a measurement or content consistency problem.
How to use it
Never report one blended number. Segment by platform, by topic or product line, and by funnel stage. A brand can lead on ChatGPT and trail on Perplexity for the same prompts, and winning awareness prompts while losing comparison prompts is a very different situation from the reverse. Promptwatch's citation share by content type report shows why intent matters: product pages take 26.0% of citations on commercial prompts but 19.1% on informational ones, so the same page strategy will not lift every segment.
Position: first name or fifth?
Position is where your brand lands in the order of brands within an answer, usually averaged over the answers where you appear. Lower is better. It matters because someone reading a list of five options rarely gives the fifth the attention of the first. OptimizeGEO makes the commercial point well: a brand mentioned first in 60% of responses is worth far more than one buried in 90% of them, because first mention reads as an implicit recommendation and a buried one reads as listed for completeness.
Two cautions.
First, check what the tool means by position. Visilay notes that Semrush measures the position of your domain's citation among the sources, not of your name in the list. Same label, two different numbers. Promptwatch also tracks citation rank separately (where your domain lands in the source list), so mention position and source position are not mixed up.
Second, single runs are noisy. Brand order frequently changes when the same prompt is run twice, so average position is only meaningful across many runs. LLM Pulse suggests 3 to 5 repeats per prompt before trusting position-weighted numbers.
A useful pattern from Visilay's test of eight ChatGPT answers: one brand had half of the leader's mentions but two first-place finishes, because ChatGPT put it on top for budget-specific questions. Another appeared often but averaged position 3.8, almost always after its rivals. Presence and prominence really are separate problems.
Share of Voice: named instead of them
Share of Voice is your slice of all brand mentions across a prompt set, compared against a defined competitor group. If a prompt cluster produces 200 brand mentions in total and 40 are yours, your Share of Voice is 20%.
Visibility Score is absolute and Share of Voice is relative. Visibility answers "how often does AI name us." Share of Voice answers "how often does AI name us instead of them," which is usually what stakeholders care about.
Three ways to calculate it
| Method | How it works | Strength | Weakness |
|---|---|---|---|
| Mention-based | Your mentions divided by all brand mentions | Simple, easy to explain | Ignores order and prominence |
| Citation-based | Your cited sources divided by all cited sources | Shows who AI trusts as a source | Misses uncited mentions |
| Position-weighted | Mentions weighted by order, often 1/n (1.0, 0.5, 0.33...) | Rewards being named first | Needs several repeats per prompt to stabilize |
LLM Pulse works through a position-weighted example: first place 10 times (weight 10.0), second 15 times (7.5), fifth 20 times (4.0) gives 21.5. Competitors score 60.0 combined, so the weighted share is 21.5 / 81.5, or 26.4%. The same brand could post a very different figure under the mention-only formula, which is why two tools rarely match.
The denominator decides everything. Use the same prompts and the same competitor list for every brand, and keep the list fixed over time. Add a competitor mid-quarter and your share drops for reasons that have nothing to do with performance. Also watch for rising mentions with falling share: that is competitive displacement, not growth.
Sentiment: how you are described
Appearing in an answer is not the same as being described well. Sentiment scores whether the model frames your brand positively, negatively or neutrally. A rising Visibility Score with declining sentiment is a warning the raw count hides: AI names you more often, but describes you worse.
Sonary points out that the word gets used two ways. Tone sentiment is about how the AI talks about you. Momentum sentiment is a looser usage where a drop in mentions gets labeled "negative." Those are different things. If mentions fall from 1,000 to 700, tone could still be glowing on every one of the remaining 700. Check tone separately before concluding that perception got worse.
When tone is negative, the driver is often a third-party source shaping how models describe you, which points the fix at reputation work off your own domain. Promptwatch scores sentiment from 0 to 100 (around 50 is neutral) independently of visibility, and pulls out the individual claims AI makes about you, such as "affordable for small teams" or "its mobile app lacks features," so you can see which specific statement is hurting you. Not every tool covers it. Sonary notes Ahrefs Brand Radar focuses on share of voice and mentions without a dedicated sentiment feature, so confirm before you build reporting around it.
What the four metrics look like side by side
| Metric | Question it answers | Typical formula | Low value means | Common trap |
|---|---|---|---|---|
| Visibility Score | Does ChatGPT name us? | Responses with brand / total responses | Coverage problem | One blended number across platforms |
| Position | How early are we named? | Average order when mentioned | Prominence problem | Confusing mention position with citation position |
| Share of Voice | Are we named instead of rivals? | Your mentions / all brand mentions | Losing the category | Changing the competitor list mid-trend |
| Sentiment | Are we described well? | Positive, neutral, negative share, or 0-100 | Reputation problem | Mixing up tone and momentum |
Reading the metrics together
The metrics are most useful as a diagnostic. A few combinations come up again and again:
- Low Visibility and low Share of Voice: a coverage problem. Your content does not match what ChatGPT searches for. Query fan-outs help here, since tracked prompts are not what the model actually searches. Promptwatch's query fanout data shows averages of 2.15 fan-outs per response in December 2025, falling to about 1.84 by March 2026, so match your pages to the real sub-queries.
- Decent Visibility but late Position: a prominence problem. You are in the consideration set but not the lead recommendation. Strengthen comparison and product content.
- Mentions without citations to your own domain: a content or crawlability problem. Check whether AI crawlers can read your pages at all.
- Rising Visibility, falling Sentiment: a reputation problem. Look at the third-party sources shaping the description.
- Stable mentions, falling Share of Voice: a competitor is gaining. Compare topic by topic.
Promptwatch's own guidance from the unique-domains report is practical here: track where you appear and how ChatGPT describes you, check which competitors are cited in the same responses, and measure across many prompts because one prompt tells you little.
Why two tools show different numbers
Beamtrace's comparison of ChatGPT visibility trackers makes the point that visibility has no single formula. Some tools lead with a Visibility Score and add rank and Share of Voice around it. Others keep Position, Sentiment and Share of Voice alongside Visibility rather than blending them. Differences also come from:
- Prompt set: your prompts versus theirs, branded versus unbranded.
- Data source: the real ChatGPT interface versus the API, which can return different answers and citations.
- Sampling: one run versus several repeats per prompt.
- Model version and location, which change answers.
- Denominator and competitor set for Share of Voice.
The practical rule: pick one tool, fix the prompt set and competitors, and compare against your own baseline. Do not compare a Visibility Score from one vendor to another vendor's.
Setting up a trustworthy measurement routine
- Build a prompt set from real buyer language across awareness, comparison and purchase stages. Include unbranded prompts, since those show whether you are discovered.
- Add personas if your buyers differ, because the same question phrased by a CFO and by a developer can pull different brands.
- Fix a competitor list of 4 to 8 brands and leave it alone.
- Run each prompt several times and average the results.
- Segment by platform, topic and intent. Zenmedia's example shows why: one brand's answer share on Claude moved from 0.10% to 7.70% while on ChatGPT it only went from 6.60% to 7.30%.
- Report trends against a baseline, monthly for leadership and weekly for the team doing the work. Zenmedia cites a Semrush finding that 45% of marketing leaders cannot accurately measure their brand's visibility in AI answers, so a clean baseline is already an advantage.
Tools that report these metrics
Most serious trackers cover at least visibility and share of voice. The differences are in depth, sentiment, and whether they help you fix anything.
Promptwatch reports Visibility Score, Share of Voice, Position and Sentiment from the real ChatGPT interface, and ties them to citations, AI crawler logs and AI-referred traffic, so you can tell whether a weak metric is a content problem or an access problem. It also has an Action Board and Content Agent for the fixing step. Pricing starts at $95/mo (check promptwatch.com/pricing), with a 7-day free trial and a free AI visibility report if you want a baseline first. For definitions, the Academy pages on Visibility Score and Share of Voice are worth a read.
Profound is an enterprise option that reports Visibility Score, Visibility Rank and Share of Voice.
Peec AI keeps Position, Sentiment and Share of Voice alongside Visibility rather than blending them, but it is monitoring-focused.
LLM Pulse is built around share of voice across several models and is worth a look if that is your main metric.
Ahrefs Brand Radar and Semrush bolt AI tracking onto existing SEO suites. Sonary notes that Brand Radar centers on share of voice, while Semrush's AI toolkit includes sentiment, at about $99/mo per domain for 25 daily prompts.
For more options, the GEO software directory at bestgeosoftware.com lists many more trackers.
| Tool | Visibility metric | Share of Voice | Sentiment | Goes beyond monitoring |
|---|---|---|---|---|
| Promptwatch | Yes, 0-100 | Yes | Yes, 0-100 plus claims | Yes (content, crawler logs, traffic) |
| Profound | Yes | Yes | Yes | Limited |
| Peec AI | Yes | Yes | Yes | No |
| LLM Pulse | Yes | Yes | Varies | Limited |
| Ahrefs Brand Radar | Mentions | Yes | No dedicated feature | No |
| Semrush AI toolkit | Yes | Yes | Yes | Limited |
Common mistakes
- Reporting a single blended score across ChatGPT, Gemini and Perplexity.
- Treating one answer as ground truth. Re-ask variance is real.
- Only tracking branded prompts, which flatters you. Discovery happens on unbranded ones.
- Changing the competitor set and then explaining the Share of Voice drop as a performance issue.
- Ignoring sentiment because visibility is climbing.
- Using referral traffic as the main KPI. Many AI answers have no clickable link at all, and ChatGPT stopped sending a referrer on October 8, 2026, so tag links or count utm_source=chatgpt.com visits.
Bottom line
Start with Visibility Score to learn whether ChatGPT names you. Add Share of Voice to see where you stand against rivals. Use Position to judge prominence and Sentiment to judge quality. Then segment, repeat, and watch the trend. If a number moves, the combination of the four tells you whether you have a coverage, prominence or reputation problem, and that is the part that decides what you do next.