A count of brand mentions is easier to trust when someone can retrace the answers behind it. Without the exact prompt, a clear date and a record of what the answer said, two checks that look comparable may describe very different situations.
Shane’s AI Visibility Tracker is a browser-based workspace for recording those observations. You run the checks in the AI products you use, then enter the results manually. The tool keeps a log and provides a CSV export; it does not query AI services or retrieve citations for you.
Here is a practical way to keep the evidence useful after the initial check.
Keep the question and the conditions together
Enter the exact question in Prompt / question rather than summarizing it. “Which accounting tools suit a freelance designer?” and “Is Example Ledger a good accounting tool?” ask different things. A brand appearing in the second answer tells you less about unprompted discovery because the question already names it.
Use the same label for each AI platform throughout a comparison. If the product shows a model or mode that matters to your research, record that detail in your supporting notes. The tracker counts distinct platform labels as entered, so inconsistent naming can make one product appear to be several different entries.
Set Check date deliberately, especially when logging an older answer. If you leave it blank, the current implementation uses the date in UTC when you save the record. That may differ from your local calendar date near midnight.
Keep other relevant conditions alongside your evidence, such as whether you started a new conversation or used follow-up questions. The goal is to make later comparisons understandable, not to imply that every answer is generated under identical conditions.
Separate three different observations
When reading an answer, distinguish what you actually saw:
- A mention means the answer names the brand. It does not by itself establish that the answer recommends it.
- A recommendation means the answer presents the brand as an option for the stated need. Record any qualifications or limitations with it.
- A citation means the answer displays a source reference. Check what that reference supports rather than treating every linked page as an endorsement.
An answer can mention a company critically, recommend a competitor, and cite an unrelated background source in the same response. A short note that separates those observations is more useful than labeling the entire answer a win or a loss.
The tracker has fields for the brand, competitors, answer type and source notes. Use those fields consistently, and keep your interpretation separate from the wording of the answer itself.
Write a compact evidence note
The current tool saves only the first 500 characters of Source citations / answer notes. Keep a fuller answer, screenshot or other permitted record in your own research files, then use the notes field for a concise summary and a reference you can find again.
For a fictional check, a note could read:
“Example Ledger appears in a three-product comparison. No product is named the overall recommendation. A source link supports a feature description. Evidence: research file 2026-11-04-A, answer 2.”
That example describes the result without claiming a ranking or business outcome. Your own note should reflect an answer you actually reviewed. If you include a source URL, open it and check that it leads to the relevant page before treating it as evidence.
Keep the most important observation near the beginning of the note. Avoid pasting a long answer into the field and assuming the whole text will remain available after saving.
Treat the mention indicator as a shortcut
The current mention indicator checks whether your notes contain the brand text, ignoring letter case. It does not read or interpret the original AI answer.
That distinction matters. A note saying “Example Ledger was not mentioned” still contains the brand name and can be marked mentioned. A note that summarizes a genuine mention without repeating the brand name can produce the opposite result.
Keep your written observations accurate, and review them against the saved evidence before reporting findings. Do not use the Your brand mentioned counter as a verified measure of reach, recommendations or citation coverage. If you need a reviewed yes-or-no classification, maintain that classification in your analysis sheet alongside the exported records.
Compare a defined set of checks
Before looking for change, choose which records belong in the comparison. For example, you could review the same five questions on the same two platforms on two different dates. Keep brand-naming prompts separate from questions that do not name the brand.
If you calculate a proportion, state the denominator and selection rule. “Mentioned in three of the ten reviewed answers in this prompt set” describes a small observed sample. It does not establish how often all users see the brand or whether a website change caused the result.
Record uncertainty rather than filling gaps with guesses. An unavailable answer, a missing source or an ambiguous reference should stay visible in your notes so a later reviewer understands what could and could not be checked.
Export with the scope in mind
Export CSV downloads ai-visibility-checks.csv with date, prompt, model, brand, mentioned, competitors and notes. The answer-type selection is not included in that export, so preserve any classification you need in your research notes or analysis sheet.
The search field narrows the visible log. It does not narrow the dashboard totals or the export, which still covers all saved records. Check the exported rows before using them for a particular reporting period or sharing them with someone else.
Records live in the browser where you entered them. Keep periodic exports and your supporting evidence somewhere appropriate for the project. Together, they give you a trail that can be reviewed rather than a number that has to be taken on faith.
Open Shane’s AI Visibility Tracker and log one answer with enough context that someone else could understand what you observed.