1. Can you take me from this number to the exact answer?
Ask the presenter to open an observation behind the chart. You should be able to see the question, the full answer and its recorded source links, with the collection date and AI surface attached.
Read the passage where your publication appears. Is it named in the prose, cited as a source, or both? If the product claims an article supports a specific statement, ask what evidence supplies that mapping. A source list alone does not establish sentence-level attribution.
A useful result: an editor can explain the observation using the saved evidence, without relying on a score definition supplied verbally during the demo.
2. Can you keep the exact cited URL and the grouped article identity?
Provide a canonical story URL and a known alternate URL, if you have one. Ask how the tool handles redirects and tracking parameters. Check that grouping does not erase the original observed URL.
Google describes redirects and canonical annotations as signals for choosing a representative URL. That does not establish how an AI visibility vendor groups its records. Ask for the vendor’s own rule and inspect a worked example. Google’s canonical guidance.
A useful result: you can count an article consistently while still seeing the link that actually appeared. A different publisher’s syndicated copy remains a different destination unless your report explicitly defines a broader grouping.
3. Can I inspect a section without losing the underlying stories?
Ask for a view of a section such as travel or personal finance, then open its constituent articles. Check whether section membership comes from URL paths, CMS metadata or a manual mapping.
Use a difficult example: an article that moved sections, or a story whose URL does not reveal its current desk. Decide which classification your newsroom needs and whether historical reports should preserve the old assignment.
A useful result: a section total reconciles to inspectable records, and exceptions have an owner. A smooth filter animation is not enough.
4. Does the product distinguish a displayed citation from other source information?
Ask the presenter to define every source-related label. “Found,” “retrieved,” “mentioned” and “cited” can describe different observations. Where the collection does not expose a field, it should stay unavailable rather than being inferred as zero.
This matters when an editor asks whether an uncited article contributed to an answer. Unless the evidence supports that conclusion, the report should leave it unresolved.
A useful result: each label has a definition tied to a field the product actually collects. An unknown remains visible in the export.
5. Can I separate an editorial change from a question-set change?
Ask to compare two periods using the same questions and AI surface. Then add a new topic and inspect what happens to the trend.
The original comparison should remain recoverable. A question about election results and a new question about voting rules may both belong to politics, but they do not test the same article demand. Collection failures also need their own status.
A useful result: the chart identifies its comparable set and separates new questions. Ask for a sample change log, including what happens when wording or collection settings change.
6. Can the export survive a handoff to another editor?
Export a small, non-sensitive sample during the demo. Have someone who did not configure it find the question, article URL and observation date. Ask them to reproduce one reported count.
Check what remains after a subscription ends and whether raw evidence requires a different plan. Ask for the retention period in writing. If the vendor cannot expose a field, decide whether the editorial job can still be done.
A useful result: the recipient can trace a claim without a live guided tour. A PDF of the dashboard may serve a presentation; it may not serve an audit.
7. Can referral traffic stay separate from sampled citations?
Bring the analytics team into this part. Visits recorded on your site and citations observed in a selected AI-answer panel come from different populations. Dividing one by the other does not create a click-through rate.
Ask the vendor to show both with their own dates and denominators. If it offers attribution or revenue estimates, require the method and the missing-data assumptions before adopting them.
Check what an impression measures
Ask what each reach or demand number counts before using it to value your coverage. Ahrefs’ metric documentation defines impressions using Google search volumes associated with prompts where a brand appears. Its AI share of voice compares those impressions across tracked brands. That calculation does not establish how many people saw your article in an AI answer. Ahrefs metric definitions.
Peec's pricing page defines prompt volume as a relative score from 1 to 5 for the topic behind each tracked prompt. A score of five is not a count of five queries or five readers. Ask how the vendor estimates demand and which period, geography and population the estimate represents. Peec pricing and metric definitions.
During the demo, have the presenter trace one reported number to its inputs. Record whether it counts observed answers, estimates demand or measures visits. Keep those units separate when comparing products or building a business case; the same label does not establish the same calculation.
Apply this check to every vendor, including Gist. Choose the product whose evidence supports your newsroom's job: inspecting citations, researching demand or measuring audience activity.
A useful result: an editor can see whether source visibility and recorded visits moved, while the report remains clear about what has not been connected at the user level.
Make the demo end with a decision
Mark each check as passed, accepted limitation or unresolved. Record the evidence you saw and the person responsible for any follow-up. If article-level inspection is essential and the product cannot provide it, a larger domain chart will not solve that gap.
Use the same checklist across your shortlist, including Gist if you are considering it. Compare what each product demonstrates against the work your editors need to finish.
Related: Compare analytics and AI visibility measures and inspect an API sample record.
Start with your publication’s current presence. Try a free Gist Quick Scan using your publication’s brand or website and a relevant topic. Use that snapshot to decide which article-level questions to bring to a fuller demo.