Key takeaways
- Low Share of Voice is usually a coverage problem: AI answers don't include you often enough. Late Position is a prominence problem: you show up, but after the competitors. Weak Sentiment is a reputation problem: you show up, but AI describes you badly.
- Visibility Score is a blended number, so it can hide which of the three problems you have. Read it last, not first.
- A mention without a citation to your own site points to a content or crawlability issue, not a brand issue.
- On brand-specific prompts the source list is short, so a single review site or forum can shape how AI describes you.
- Every share-of-voice number depends on its denominator. Write down the competitor set, prompts, models, and dates next to the percentage.
Why one number is never enough
Most teams start AI visibility work by looking at a single score. It goes up, everyone is happy. It goes down, someone panics. Neither reaction tells you what to do on Monday.
The trouble is that a blended score can fall for very different reasons. You might have dropped out of answers entirely. You might still be there but listed fifth. Or AI might have picked up a bad claim about your pricing and started repeating it. Those are three different problems with three different fixes, and a content team that treats them the same wastes weeks.
This guide treats the four metrics as a diagnostic set. You read them together, work out which of the three problems you have, and then pick the fix. The same logic applies whichever tool you use to collect the data.
The four metrics, defined plainly
Visibility Score
Visibility Score is a composite. In Promptwatch, each AI response is scored on placement, share of attention, and depth, then the scores are averaged on a 0-100 scale. Responses that don't mention your brand count as zero. That last detail matters: a brand that appears in 30% of answers cannot score above 30 on a clean reading, however well it does when it shows up.
Other vendors build their own versions. OptimizeGEO describes its score as a weighted mix that includes prompt coverage (20%), answer prominence (15%), and sentiment (about 15%). HubSpot's AEO tooling frames brand visibility score as the share of tracked prompts that returned a mention, with sentiment and citation data layered on. Two tools can score the same brand differently, so never compare a number from one vendor against a number from another. For the Promptwatch definition, the Visibility Score academy page is the reference.
Share of Voice
Share of Voice is your slice of all brand mentions in a tracked set. In Promptwatch it counts one mention per response per brand. The formula is the same one marketers have used for decades: your metric divided by the total across your competitor set, times 100. What changes is what you plug in.
That is the part people forget. GrowthX puts it well: change the competitor set and the percentage changes without any change in your own activity. PRLab makes a similar point about channels. A brand can gain earned-media share of voice while losing SEO share against the same three competitors, and both readings are accurate because they describe different pools of attention.
The Share of Voice academy page covers the AI-specific version.
Position
Position is where your brand lands in the order of brands within an answer. Lower is better. First mention works as an implicit recommendation. A brand named first in 60% of responses is worth far more than one buried at the end of 90% of them, which is the example OptimizeGEO uses and one I agree with.
Sentiment
Sentiment is scored from 0 to 100, with roughly 50 neutral, and it is scored independently of visibility. That independence is the whole point. A crisis can push your visibility up while damaging demand, as the GrowthX guide notes about raw share of voice: negative, neutral, and positive coverage all count together unless you split them.
The diagnostic matrix
Read the metrics in this order and the problem usually reveals itself.
| Pattern you see | Likely problem | What it means |
|---|---|---|
| Low Share of Voice, low Visibility Score, few responses mention you | Coverage | AI doesn't include you for the prompts that matter |
| Decent Share of Voice, late Position | Prominence | You are listed, but competitors are framed as the default pick |
| Good Share of Voice and Position, low Sentiment | Reputation | AI describes you with weak or negative claims |
| Mentions but few citations to your own domain | Content or crawlability | AI knows you exist but doesn't source you |
| High Share of Voice that spiked suddenly, Sentiment falling | Possible crisis volume | Visibility is up for the wrong reasons |
Notice that Visibility Score doesn't decide anything on its own. It confirms what the other three say.
Diagnosing a coverage problem
Coverage means the prompts your buyers ask never produce your name. Symptoms: low Share of Voice, a Visibility Score dragged down by lots of zero-mention responses, and a topic view where whole themes are blank.
One reason this is common: organic answers carry a lot of sources. In Promptwatch Data's report on unique domains per ChatGPT response (Sep 28 to Oct 4, 2026), 52.5% of organic prompts cited 10 or more domains, and 47.6% of all responses did. Being absent is normal. Being the only brand present is rare. Competition for a slot is real, and a single prompt tells you little, so measure across many.
What to check
- Which prompts and topics return zero mentions, grouped by topic rather than one by one.
- Which competitors appear in those same responses.
- Which pages AI cites instead of yours, and what those pages cover that yours don't.
- Whether AI crawlers can read your site at all. If GPTBot or ClaudeBot hit errors, no amount of new content helps.
- Whether you have a page for the prompt. A content gap analysis maps your pages against the prompts AI answers.
The usual fixes are new or refreshed pages written against the exact questions, plus removing crawl blocks. Tools in the AI visibility category differ a lot on whether they help with that second half. Promptwatch's content gap analysis and AI crawler logs cover both the page question and the access question. For wider options, the GEO software directory at bestgeosoftware.com lists alternatives.
Diagnosing a prominence problem
Prominence is the quiet one. Your Share of Voice looks acceptable, so nobody worries. But Position is late, and citation rank puts your domain near the bottom of the source list.
When many sources share an answer, where you sit inside it counts as much as whether you appear. Promptwatch's own advice from the domains report is to track where you appear in the response and how ChatGPT describes you, not just whether you're cited.
What to check
- Position by prompt type. Comparison prompts ("X vs Y") often put the brand named first in the question first in the answer, so late Position there may be structural.
- Which competitors consistently sit above you, and on which topics.
- Your citation rank: closer to 1 means a primary source, further down means a supporting mention.
- Whether your page answers the question directly, early, and with specifics. Pages that bury the answer under a long intro rarely become the lead source.
Fixes are mostly editorial: tighter answers near the top of the page, stronger proprietary data, clearer product and pricing pages. Product pages are the most cited content type in ChatGPT at 22.4% of citations (Sep 28 to Oct 4, 2026, per Promptwatch Data), so thin product pages are an obvious place to start.
Diagnosing a reputation problem
Sentiment is where a decent-looking dashboard can mislead you. A brand with 30% citation frequency but consistently negative framing may convert worse from AI-referred traffic than a competitor at 20% with positive framing. OptimizeGEO makes the same point.
The key move is to stop looking at the average and look at the claims. Promptwatch's Brand Narrative pulls out the individual statements AI makes about you ("affordable for small teams", "mobile app lacks features"), groups repeats, and files them under themes like Pricing, Features, Support, or Reputation. Handraise argues something similar for PR: narratives are a better unit of measurement than individual articles.
Where reputation problems come from
On brand-specific prompts the source list is short. In the same Promptwatch Data report, 5 domains was the most common count for prompts like "is Asana good for small teams", and only 21.3% of those responses cited 10 or more. That means each source has outsized influence. A single outdated review page or a negative forum thread can dominate the description.
What to check
- Which sources feed the negative claims. Check the review sites and forums on the short source list.
- Whether the claim is simply wrong. Mark it inaccurate, then find the page on your own site that states it or its opposite. Sometimes your own old pricing page is the culprit.
- Sentiment by persona. A buyer-type persona may get a very different picture than the general audience.
- Sentiment by competitor, to see whether the problem is category-wide or yours alone.
Fixing reputation is slower. You correct your own pages, update third-party profiles, earn better reviews, and publish evidence that contradicts the claim. A reputation problem is the one case where more content rarely helps. OptimizeGEO says as much: if mentions are high but sentiment is neutral or negative, the fix is better social proof rather than more pages.
When the numbers disagree with each other
Some patterns are worth memorizing.
High mentions, low citations
AI knows the brand from training data or third-party pages but doesn't source your site. That points to content or crawlability, not awareness. Check crawler logs first. An access problem and a content problem look identical in a visibility dashboard and need opposite fixes.
Share of Voice spiked overnight
Before celebrating, split it by sentiment and by source. PRLab's warning applies directly: a spike can point to a crisis as easily as a win. Duplicate coverage and a single viral thread also inflate numbers. Check whether branded search and traffic moved the same way.
Visibility Score flat while Share of Voice falls
You are holding your own presence but competitors are growing faster. This is a competitive problem, not a content one. Look at who gained and on which topics.
Sudden platform-wide drops
Check whether it is you or the model. ChatGPT's citations per response dropped around the GPT-5.3 rollout on March 4, 2026 (Promptwatch Data), and reddit.com's share of ChatGPT citations fell from roughly 4% to 0.5% on August 14, 2026 (Promptwatch Data). If every brand in your category moved together, it's a platform change. If only you moved, it's yours.
A weekly diagnosis routine
Here is a routine that takes under an hour once your monitors are set up.
- Pick one monitor, one country, one language. Mixing markets muddies every number.
- Look at Share of Voice by topic. Note topics under your fair share.
- Look at Position for the topics where you do appear. Note the ones where you rank behind two or more competitors.
- Read Sentiment by theme, then open the underlying claims for anything below neutral.
- Check self-citation rate. If it is low where mentions are high, look at crawlers and page structure.
- Only then open Visibility Score to confirm the direction.
- Write one action per problem type. Do not try to fix all three at once.
Also write down the setup next to every figure: prompt set, models, country, competitor list, date range. GrowthX's advice to put the denominator beside the result is the single cheapest habit that prevents misleading reports.
Choosing tools for the job
No tool diagnoses these for you out of the box if it only reports one blended score. Look for per-model, per-topic, and per-competitor breakdowns, claim-level sentiment, and a way to act on what you find.
| Tool | Strongest at | Gap to watch |
|---|---|---|
| Promptwatch | Metrics split by coverage, prominence, and sentiment, plus crawler logs, content gaps, and content creation | Content Agent and faster tracking need the Business plan |
| Profound | Enterprise monitoring across engines | Higher price point, less execution help |
| Peec AI | Simple multi-language prompt tracking | No crawler logs or content generation |
| Otterly.AI | Affordable monitoring | Basic tracking only |
| HubSpot AEO | Fits inside an existing HubSpot workflow | Narrower AI-specific data |
| Semrush One | Adds AI tracking to an SEO suite | Fixed prompt sets in many setups |
For a wider list, the AI rank tracking directory at ai-rank-tools.com is a good place to compare more options. If you want a quick baseline before committing, Promptwatch has a free AI visibility report and a 7-day trial.
Making the diagnosis stick with stakeholders
The hard part is often not the analysis. It is explaining why a high Visibility Score is not good news yet.
A few things help. Report Share of Voice with its denominator. Show Sentiment beside it, always. Frame each finding as one of the three problems with one owner: content team for coverage, SEO and editorial for prominence, comms and product marketing for reputation. Then tie the fix to a prompt set you will re-measure in two to four weeks. A recommendation without a way to verify it tends to turn into a debate.
One caution, borrowed from PRLab. A stronger share of voice does not by itself prove more sales or leads. Connect the visibility numbers to AI-referred traffic and conversions separately, using AI traffic attribution, before claiming revenue impact.
What to do first
Start with whichever of the three problems the data points to, and resist the urge to fix everything. Coverage problems get new pages and crawl fixes. Prominence problems get sharper, more specific pages and better sourcing. Reputation problems get claim-level investigation and corrections at the source. Then re-run the same prompts and check whether the right metric moved. If Share of Voice rose but Position didn't, you solved a different problem than the one you had.