Measure and understand
AI Visibility
See how your brand and competitors appear in the answers your project measures.
Use AI Visibility to investigate brand presence across a defined set of prompts and engines. Start with the selected run and filters, then move from the summary into the answer evidence.
Run a useful audit
Choose the prompts, engines and repetitions in the audit launcher. CiteLadder measures answers from ChatGPT, Claude, Gemini and Google AI Overviews, collected from the consumer apps and from provider APIs; a collected answer can differ from what a signed-in user sees. Review the estimate before confirming. More repetitions give more observations under that run’s conditions; they do not turn a sample into a guarantee about every user’s experience.
An audit keeps the prompts, tracked roster and measurement configuration it used. Check Runs for completion, cancellation and unsuccessful attempts.
Measure more than one market
A project measures from its own country and language by default. On Growth and Scale plans you can add markets, a country with a language, in the project edit panel under Additional markets (Growth includes 3, Scale 10). When a project has more than one market, the audit launcher and Scheduled audits ask which markets to measure; each market becomes its own run, launched together.
Answers collected from the consumer apps (ChatGPT Search, Gemini and Google AI Overviews) can only be measured in the countries and languages that collection supports, which differ by app; a regional language such as Brazilian Portuguese is measured as Portuguese, and the launcher refuses an unsupported pick before anything runs. The Gemini API has no location setting, so it runs only in the default market and shows as not available in others. A prompt that names a place is still measured from the market you pick.
Visibility then shows a market switcher, and Trends opens with By market: each market’s latest run side by side, with the change since that market’s previous comparable run. Markets are never pooled, and a market you have not run yet shows as not run, not zero.
Schedule audits
To measure on a regular cadence, open Runs and use Scheduled audits. Choose a prompt set, the measurement scope, the engines and a cadence: one time, every few minutes (at least five), hourly, daily or weekly. The first scheduled run starts right away, and later runs follow your browser’s time zone.
A schedule pauses itself when it cannot start runs, for example when your plan or trial has ended, an engine connection is missing or the included budget is used up. The row shows why it paused; fix the cause and choose Resume.
Read Trends
Trends is the default view. Select the run or period, engine and cohort appropriate to your question. When using a baseline, check whether the comparison is compatible.
Trends reports four measures. Keep them separate:
| Measure | What it tells you |
|---|---|
| Visibility | The share of answers that name your brand (the mention rate) |
| Share of voice | Your share of all mentions of the brands you track; repeated mentions in one answer count once |
| Owned citation rate | The share of answers that link to a page you own |
| Average position | Where your brand tends to appear among the brands you track, counted only in answers that name you |
A citation is not a brand mention, and a mention is not an endorsement. Average position ranks you among your tracked brands, not within the whole answer. Measures a run cannot support remain unavailable rather than zero.
Read Perception
Perception shows how answers portray your brand and your competitors. Every answer from a brand audit that names you or a competitor is read for sentiment, at no credit cost. Answers from audits run before this view existed are not classified.
- Net sentiment runs from −100 to +100: positive mentions minus negative mentions, divided by every classified mention. Mixed mentions count in the denominator only.
- Coverage always sits beside it as “N of M mentions classified”. Mentions still being classified, mentions the classifier could not judge, low-confidence labels and mentions left unclassified (for example when the platform limit is reached) are counted separately and never treated as zero.
- Themes group what answers praised or criticised (pricing, support, reliability and so on), each with a quote.
- Criticism in answers lists the negative points about you, quoted exactly as the answer put them. Open a quote to see the full answer.
- Sources cited alongside criticism lists domains cited in answers that criticised you. They appeared alongside the criticism; the view does not show that they caused it.
- Recommended is the share of answers naming you that explicitly recommend you near your first mention. It reads English phrasing only.
When nothing is classified yet the view says so: answers are still being classified, perception is unavailable (with the reason), or no answer in the selection named you. A change in how perception is classified starts a new comparison series, so points from before and after it are not compared directly.
Check facts in answers (pilot)
In workspaces in the fact-checking pilot, Agent → Context lists Brand facts: short statements such as a plan’s price, an integration or the markets you serve. A new fact starts as a draft; only confirmed facts are used. Audits started after you confirm a fact check the factual claims answers make about you against the facts confirmed at that moment, at no credit cost. Editing a fact later never changes an earlier run.
The Accuracy tab shows the result:
- Accuracy is supported claims divided by claims that were supported or contradicted. Coverage sits beside it; claims still being checked, low-confidence claims and claims that could not be checked are counted separately and never treated as zero.
- Contradicted claims quote what the answer said, next to the fact it contradicts. Open a claim to see the full answer.
- Not covered means no confirmed fact is about that point, and inconclusive means a related fact neither confirms nor rules it out. Neither is an error in the answer.
- Sources cited alongside contradictions lists domains cited in those answers. They appeared alongside the claim; the view does not show that they were its source.
Opinions and praise are not checked here; they appear in Perception. A change to your confirmed facts or to how claims are checked starts a new comparison series.
Ads in ChatGPT answers
Ads shows the paid ads that appeared in ChatGPT Search answers in the selected runs. Ads are reported on their own: they never count as citations or sources and never change mention rate, share of voice or scores.
- Ad presence is the share of ChatGPT Search answers that showed at least one ad, written as “N of M ChatGPT Search answers”.
- Your ad share is your share of every ad appearance, with your best ad position. When none of your ads was seen it reads “Not advertising”, not 0%.
- Advertisers lists who advertised, whether each is you, a competitor or another business, how often and on how many prompts, and when they were first and last seen.
- Prompts that surface ads lists the prompts whose answers showed ads, the top advertiser, and how often the answer still mentioned you when a competitor advertised. These are counts, not a cause.
- Ad creatives lists each ad as text with the page it leads to. Tracking parameters are removed and ad images are not shown.
Only ChatGPT Search shows ads. Other surfaces read Not applicable, never zero. ChatGPT Search answers collected before ads were read show ads as unavailable; ads appear from the next audit. Ads are those seen in ChatGPT sessions collected for your project’s market: what your buyers see depends on their plan, account and country.
Each ChatGPT Search answer in a run also lists its ads under the answer, separate from its citations.
Inspect Sources
Switch to Sources to see the domains and URLs used in the selected evidence. Open a domain to inspect its URLs and the prompts that reached it. Open an answer to check how a citation was used.
See Sources and citations for interpreting retrieval, citation rates and brand co-occurrence.
Read Query Fanout in context
Query Fanout shows captured query evidence associated with the measured answers. Use it to understand the recorded retrieval context. Absence of captured fanout does not prove that an engine did no research.
Respond to a change
First check whether the engine, prompt set, cohort or successful-answer coverage changed. Then inspect which answers and sources explain the observed difference.
Use an Action or a focused Agent question to turn the evidence into work. Avoid attributing movement to a website edit until you have examined later compatible measurements and their limitations.