FAQ

Scores

My score changed and I didn’t do anything. Why?

Because the thing being measured changed. AI assistants re-retrieve constantly: a new article gets indexed, a competitor publishes, a source drops out of the retrieval set, and the same question gets a different answer. Model-side updates move things too.

Treat a single run as a snapshot with real variance, and the trend over several runs as the signal. If you want a movement to be attributable, keep the prompt selection, engines and markets identical between runs — changing the input changes what the trend means.

Why is my Mention Rate lower than the number of answers that mention me?

Mention Rate counts discovery prompts only — the ones that never name your brand. Comparison and trust prompts guarantee a mention, so counting them would push the number toward 100% as you add more of them. Their coverage is reported on the second line of the card as x/y in brand-named prompts. → Prompt intents

Why is my Sentiment high when my Mention Rate is low?

Sentiment is averaged over answers that actually mention you. If you’re mentioned rarely but described well when you are, that is exactly what the two numbers should show. Fix visibility first; sentiment is only actionable once there’s something to be sentimental about.

A probe shows — instead of scores. What happened?

The analysis call failed after retries — a timeout, a rate limit, or an unparseable response. The answer text and citations are still stored, but the probe is excluded from every metric, including the mention-rate denominator, because “not measured” is not the same as “performed badly”.

There’s deliberately no keyword fallback guessing whether you were mentioned: string matching can’t tell your brand from a same-named company, from an ordinary word, or from a hit inside an echoed search URL — and all three errors overstate visibility. → Averages & denominators

Why does my Final Score differ from my AI Score?

The Final Score blends in the technical health of your own site: 0.8 × average AI Score + 0.2 × SEO Score. A large gap between the two means your site is dragging — check the Site Audit. If no Lighthouse audit exists, the two numbers are identical, because the SEO term is dropped rather than counted as zero. → Final Score

The AI mentioned a company with my brand’s name, but it isn’t us. Is that counted?

It shouldn’t be. The grader is given your brand name and your website URL specifically to disambiguate, and is instructed that a same-named company in a different industry is not a mention. If you see a case where it got this wrong, the probe detail page shows the exact answer it judged — that’s the fastest way to confirm.

Regions and engines

Why do two regions show the same score?

Check which half is identical. The SEO Score is site-level and cannot vary by region — your site is the same site everywhere — so region cards mark it site-level. If the AI halves are identical too, the engines genuinely returned equivalent answers in both markets, which happens for global brands and English-language queries.

Why isn’t Claude in my region comparison?

Because it has no geographic signal to compare. Cross-region tables use only engines whose retrieval actually runs from the target country, so every row has the same prompts and the same engines behind it. Claude appears under Global engines, and Doubao appears under its home market. → Engines & regions

Adding a region didn’t double my probe count. Is that a bug?

No. Region-aware engines run once per market; Claude and Doubao run once in the primary market whatever the market count. With ChatGPT + Gemini + Claude and two markets, 10 prompts costs 10 × (2+2+1) = 50, not 60. → Plans & usage

Prompts

The intent on my imported prompt is wrong. Does it matter?

Yes — the intent decides which rubric grades the answer and which averages the probe counts toward. A branded prompt sitting in the discovery bucket inflates Mention Rate; a discovery prompt in the branded bucket loses its position score. Fix it in the Prompts tab.

Why can’t my prompts name a country?

Because the market is a parameter of the run, applied at retrieval time. A prompt containing “in the UK” would apply the market twice — once in the text, once in the retrieval — and its results could no longer be compared across markets.

How many prompts do I need?

Ten shows a signal. A few dozen makes the per-engine and per-region splits meaningful. Beyond that, breadth of intent matters more than raw count: category, scenario, comparison and trust each answer a different question.

Improving the score

What actually moves AI visibility?

In rough order of leverage:

1. Be reachable and readable If AI crawlers are blocked or your content only exists after JavaScript runs, nothing else matters. Start at the Site Audit.
2. Get onto the sources engines already read Look at cross-model cited sources. A page several engines read is a page where one change propagates widely.
3. Publish pages that answer the actual questions Engines cite pages that answer prompts. A site with no such pages has nothing to be cited for.
4. Win the long tail before the head Scenario prompts are winnable and closest to a buying decision; category head terms are incumbent territory.

How long before a change shows up?

On the retrieval cycle of each engine, which varies — days to weeks. A new page has to be indexed before it can be retrieved, and retrieved before it can be cited. Run on a stable cadence and read the trend rather than the next run.

Should I re-run after a failed batch?

Only if a large share of probes failed analysis. The panel reports how many were excluded; a couple of failures out of dozens changes nothing meaningful, and a re-run spends allowance for the same answer.