AI Visibility Repeated-Run Log Template

AI Visibility Repeated-Run Log (fill-in)

When to use: One row per run of a prompt, with the model, date, run number, the sources the answer cited and a short variability note, plus a completion check.

Repeated-run log

One row per run. Keep every run of the same prompt together, so you can see how the answer moved. Record the row right after you read the answer.

Prompt tracked: [the exact question, word for word] - Log owner: [name] - Runs per round: [number]

Per-run detail (fill for every row)

• Brand named: [yes / no], and where in the answer it sat
• Your page cited: [yes, which URL / no]
• Conditions: [logged out, no history, country, language]
• Answer gist: [one line on what the assistant recommended]
• Screenshot saved: [file name or link]
• Notes: [follow-up question asked, any refusal, anything odd]

Next action

• Log owner: [name] - task: run the prompt the agreed number of times and fill a row each, by [date]
• Content owner: [name] - task: review runs where the brand was missed and plan the fix, by [date]
• Reviewer: [name] - task: tally the citation rate and set next round, every [weekday]

Completion check

• Every run has a model, a date with time, a run number and the cited sources.
• Runs of one prompt share the same conditions, written in the detail list.
• Each row says whether the brand and the page were cited.
• Rows are grouped by prompt, so a reader can see run 1 to run N in order.
• A citation rate for the round is recorded once all runs are in.

What to Log Each Run (field guide)

When to use: Plain definitions for every field, so two people log the same run the same way.

Field guide

Fill these the same way every time. A log only tells you something when the runs are comparable.

Recording cited sources well

• List sources top to bottom, as the answer shows them.
• Mark your own domain clearly, even when it sits last.
• Note a brand named in the text but not linked. That still counts as a mention.
• Keep a screenshot, so a disputed run can be checked later.

To compare with your own runs, look up your domain in the AI Visibility Checker: it shows recorded AI answers that name or link you, with every cited source.

Run Cadence & Sampling Rules

When to use: Decide how many runs per prompt and how often to repeat a round, so the numbers mean something.

Cadence and sampling

Set these once per prompt and write them in the log header. Then every round is run the same way.

Why repeats are the point

• The same prompt can return different text even at the most deterministic setting. Researchers showed this across several models.
• Ahrefs tracked Google AI Overviews and found the answer and its citations for the same query change often between checks.
• So one screenshot is a coin flip. A round of runs is a measurement.

Sources: the arXiv paper Non-Determinism of Deterministic LLM Settings and Ahrefs' study of AI Overview change, both checked 13 November 2026.

Variability Scoring Guide

When to use: Turn a round of runs into a citation rate and a plain variability label you can report.

Scoring a round

When all the runs for a prompt are in, work out two things: how often your brand was cited, and how steady the answer was.

Citation rate

• Count the runs that cited your page, or named your brand, out of the runs in the round.
• Write it as a plain fraction, such as 2 of 5, next to the prompt.
• Keep brand mentions and page citations as two separate counts. They are not the same win.

Variability label

• Stable: the same sources appear in nearly every run, in a similar order.
• Mixed: your brand shows up in some runs and not others.
• Volatile: the cited sources change a lot from run to run, with little overlap.

What to do with each label

• Stable and cited: hold the page, watch for slippage each round.
• Mixed: the opening exists. Strengthen the page so more runs pick it.
• Volatile or never cited: treat it as a gap. Plan content for the question.

When a prompt is mixed or missing you, open your page in the Content Optimizer and compare it with the pages ranking for that topic.

Spreadsheet Header (CSV)

When to use: A paste-ready header row and allowed values to run the log in Google Sheets or Excel.

Paste-ready spreadsheet header

Paste this row into cell A1 of Google Sheets or Excel, then split text to columns by comma. One row per run.

prompt_id,prompt_text,model,run_datetime,time_zone,run_number,runs_in_round,conditions,cited_sources,brand_named,page_cited,page_url,variability,citation_rate,screenshot,owner,notes

Allowed values

• brand_named: yes, no
• page_cited: yes, no
• variability: stable, mixed, volatile
• conditions: logged_out, history_off, plus country and language codes

Sheet tips

• Group rows by prompt_id, then sort by run_number, so a round reads in order.
• Fill citation_rate once per round, on the last run row, as a plain fraction.
• Highlight runs where page_cited is no but a competitor appears every time.
• Keep old rounds. The trend across weeks matters more than any one run.

Find question ideas for your prompt list in the Questions tab of Keyword Research, then log a round for each.

Weekly Review

When to use: A short weekly routine for the reviewer: new rounds, citation rates, volatile prompts and content gaps.

Weekly review

Pick one day a week. The reviewer reads the latest rounds and turns them into next steps.

Checks

• Coverage: every tracked prompt has a full round this week, run the same way.
• Citation rate: compare this round to last for each prompt. Note any that fell.
• Volatile prompts: flag prompts where the sources swing a lot; they need a stronger page.
• Never cited: list prompts where no run named you. These are the content gaps.
• Wins: prompts that moved from mixed to stable after a change.

Review fields

• Review date: [YYYY-MM-DD] - Reviewer: [name]
• Prompts up this week: [list]
• Prompts down or volatile: [list]
• Gaps to brief: [prompts with no citation]
• Next round date: [YYYY-MM-DD]

For prompts where you are never cited, the AI Content Brief Template turns a gap into a page, and a full programme is covered by SEO and GEO services.

Worked Example (fictional Vellcrest proposal software)

When to use: One round of five runs for an invented proposal-software brand: cited twice, missed three times, with a variability read.

Worked example: Vellcrest

Example only. Vellcrest is an invented proposal-software brand. The prompt, models, sources, dates and counts are made up to show a filled-in round.

Prompt tracked: best proposal software for small agencies - Log owner: content lead - Runs per round: 5

How the round was read

• Citation rate: Vellcrest cited in 2 of 5 runs. Brand named, page not linked, in both.
• Variability label: mixed. The brand appears and disappears across runs.
• Steady rival: rivaltool.example showed in every run, so it owns this answer.

Why this works

• Five runs, same conditions, so the 2 of 5 is a measurement, not a guess.
• Grouping by prompt made the swing obvious at a glance.
• The gap is now clear: write toward what rivaltool.example covers that Vellcrest does not.

Want help turning repeated-run data into a GEO plan for your pages? Book a free consultation with Anatolii Ulitovskyi.