How to write a weekly AI visibility report
One memo, one week, the assistants you actually queried. End with a single URL and an owner, not a tour of the category.
Built by StyleLine FZE in Sharjah. Email support@mentionclock.com. We reply within one business day.
Choosing a tool is a different job from a single method. The dated vendor scorecard is best ChatGPT brand monitoring tools.
A weekly AI visibility report is a short memo that says what named your brand in assistant answers, what changed since the previous run, and what one person will do next. It is for a boss who will not open a dashboard, or for a client who forwards a single email. It is not a market share study, and it is not a promise that a model will recommend you.
Write it from a frozen prompt list and from runs you can date. Mention Clock records mention, place in the answer, rivals, share of mentions, tone, accuracy flags, and cited pages. Free and Starter run weekly. Pro and Agency run every 3 days, so the memo can still go out once a week and use the latest completed run. Every plan sends a Monday recap and a mail when you drop off a prompt you used to win. Those mails are notes. The memo is the judgment you add on top.
Who the memo is for
Write one memo per brand. An agency should not blend two clients into one narrative. Mention Clock keeps a project as one brand. Agency, at $199 a month or $1,990 a year, is the plan with 15 projects, client workspaces, and a white label PDF you can attach instead of pasting tables into a slide. A single in house brand can stay on Free, which is $0 for 1 project, 5 prompts, and 2 engines, or on Starter at $29 a month ($290 for the year) for 25 prompts. Pro at $79 a month ($790 for the year) fits a longer list and the cited source report. If the reader is a founder with five questions, do not dress the memo up as an agency readout. Match the length to the prompt count they can recognize.
Say who asked. "For the Thursday sales standup" and "for the retainer recap" want the same facts and a different first sentence. The standup wants the prompt that dropped. The retainer wants the action that will be on next month's invoice. Both still need the engines named in the header so nobody thinks you measured every assistant on the internet.
Header block
Put these lines at the top, before any chart:
- Brand and domain
- Week ending date, in the timezone you will repeat next time
- Engines included, by name
- Prompt count, and a yes or no on whether any wording changed
- Rival names stored for the share figure, or "none" if you did not track rivals
On a paid plan the engines you can name today are ChatGPT, the second chat engine, and Perplexity. On Free, name the two you selected. The upcoming engine is being added, so leave it out of the header until a run actually includes it. If someone checked an AI Overview by hand, label that row as a manual Search note. Mention Clock does not scrape Overviews, and the memo should not imply that it does. A reader who mixes those columns will brief the wrong team.
If you changed a prompt, say so in the header and start that prompt's history over. A renamed question is not a continuation. People forgive a short list. They do not forgive a trend line that secretly compares two different questions.
Body, in an order a reader can skim
Coverage. One sentence on what you asked. Example shape: "12 buyer prompts, unchanged, on ChatGPT and Perplexity." Do not paste every prompt into the mail if the list is long. Link the list or attach it. Do paste any prompt that dropped, because that wording is the evidence.
Mentions. For each engine, say how many prompts named the brand. Keep the engines in separate sentences. A brand can lead in the second chat engine and miss Perplexity on the same Tuesday. Averaging them into one score hides the action. Place, when you were named, can be "first among the brands in that answer" or "later in the list." You do not need a fake rank index.
Share of mentions. If you track rivals, include the figure Mention Clock stores: your brand mentions divided by mentions of every tracked brand on that prompt list. Count each time a name appears, not once per prompt. Point the curious reader at share of voice in AI search, including the worked example from the run dated 28 Oct 2026. Use the stored figure. There is no "good" percentage in the product. The comparison is you versus the rivals on these prompts, this week versus last week.
Movement. Three bullets at most: a prompt you lost, a rival that appeared for the first time, and a cited host that was new. If nothing moved, write "no change on the frozen list" and stop. A quiet week is a valid report. Filling it with screenshots of stable answers teaches the reader to ignore you.
Facts that are wrong. Tone can be positive, neutral, or negative. Accuracy flags cover a stale price or a dead feature. Quote the wrong claim and the prompt it appeared on. Do not argue with the model in the memo. Name the page that should be corrected, because that is the lever you have.
Cited hosts. List hosts, not a dump of every URL, unless a single path is the task. Pro and Agency gather this in the cited source report. On Free and Starter the URLs are still on each answer, and you can roll them up yourself. Call out hosts that mention rivals and skip you. That line is more useful than a count of total links.
One action. Owner, URL, and the prompt it should affect. "Rewrite the comparison that still shows the retired tier" is an action. "Improve AI visibility" is not. If you have two actions, pick the one tied to a prompt the reader cares about and park the other in a backlog the memo does not pretend to manage.
What the reader should be able to repeat
A good memo can be forwarded without you on the thread. After one reading, someone else should be able to say: which assistants, which week, whether the questions stayed still, whether you were named, who else was named, which site got the citation, and what will be edited. If they cannot say that, the memo is a tour.
Keep numbers you did not measure out of the mail. Do not estimate how many buyers saw the answer. Do not convert a mention into pipeline. Do not borrow a statistic from a vendor homepage. The public prices for Mention Clock are the ones on pricing. Competitor prices, if you must include them, belong in a separate note that cites AI visibility tool pricing and the date on that table, not in the visibility memo where they will be mistaken for your results.
Cadence and raw material
Decide the send time and keep it. A Monday memo matches the Monday recap mail, which is on for every plan. If your reader lives in another weekday, send then, but say "week ending" so the date is unambiguous. For a launch, Pro's every 3 days run can feed a midweek note that only lists prompts whose cited hosts changed. Do not upgrade the cadence in the memo's tone. "Every 3 days" is the product clock. Calling it a daily tracker would be false.
Raw material, in the order you should open it:
- The latest run for the project, one engine at a time.
- The drop out mail, if one arrived, because it already names the prompt you lost.
- The cited URLs on the prompts in that mail.
- Last week's memo, so you only describe movement.
If you are still in a spreadsheet, the same columns exist on the tracker file linked from how to track brand mentions in AI chats. The memo does not require the sheet once a project is running. The sheet is there for a week when you want the rows in email without a login.
Slack or Teams posts start on Starter. They are a ping, not the memo. Paste a three line summary into the channel and keep the full note in the mail or the PDF the client expects. A channel scroll is a poor archive.
Voice that survives a forward
Use the brand's name, the rival's name, and the host. Avoid adjectives about how alarming the week was. The reader can see that a prompt went from named to absent. Tell them the prompt text and the new host. If tone flipped from neutral to negative, quote the sentence. If the answer is positive and the citation is a stale review, say both, because the team that updates the review is not the team that rewrites the homepage.
For a client, attach the white label PDF on Agency and put two sentences above it: what changed, and the action you will take under the retainer. For a boss, those two sentences can be the whole mail, with the header block under them for anyone who wants the scope. Long memos get skimmed at the first adjective. Short memos get forwarded.
Do not write "we will get the brand into ChatGPT." Write "we will measure these prompts and edit the page the answers cite." That is a job you can finish. The measurement is the report. The edit is the work that might show up on a later run, and it might not show up on the very next one. Say that once so the reader does not treat next Monday as a verdict on the edit.
A worked outline you can copy into a doc
Use this skeleton and replace every bracket. Do not add sections you cannot fill from the run.
Week ending [date]. Brand [name], domain [host]. Engines: [list]. Prompts: [count], wording unchanged: [yes or no]. Rivals on the share figure: [names or none].
ChatGPT named us on [count] prompts. The second chat engine named us on [count]. Perplexity named us on [count]. Drop the lines for engines you did not run.
Share of mentions, same rival set as last week: [figure, or "not tracked"]. Definition stays your names divided by all tracked names.
Lost: [prompt text] on [engine]. New rival: [name] on [prompt]. New host: [host] on [prompt].
Wrong claim: "[quote]" on [prompt]. Page to correct: [URL].
Action: [person] will [edit or outreach] at [URL] before [date], aimed at [prompt].
That skeleton is complete. A client can forward it. A boss can read it in a minute. If a line has no fact, delete the line. An empty "new rival" line looks like you forgot, which is worse than a memo that only has coverage and an action.
Where screenshots fit
Sometimes the week starts because sales forwarded an image. Treat the image as a lead, not as the result. Recover the prompt, confirm the product in the corner of the image, and put that prompt on the frozen list if it is a real buyer question. The method is how to turn a ChatGPT screenshot into a weekly check. The memo then reports the scheduled runs, and it can mention the image in one clause: "Sales forwarded a ChatGPT session on [date]. We added that prompt. The figures above are the later runs, not the image."
Personalized apps and API runs can differ. Say so if a stakeholder waves a phone that disagrees with the table. Direction across repeated runs is what the memo stands on. One session on a logged in phone is not a rebuttal of a month of the same wording, and a month of API rows is not a rebuttal of a specific buyer's screen either. Put both in the notes when they conflict, and do not average them.
Closing the loop next week
Open next week's draft by copying this week's action line to the top and marking it done, slipped, or blocked. Then write the new coverage from the new run. Readers trust a report that remembers its own homework. They stop reading a report that rediscovers the category every Monday.
If the action was a page edit, look at the cited hosts on that prompt before you claim credit. The host may be unchanged. Report the unchanged host. The edit can still have been the right page to fix. Credit language that outruns the run will make the next drop look like a failure of the whole program.
A free project is enough to practice the memo for a month: create a free project. Five prompts, two engines, weekly, no card. Move up only when the reader asks for the third engine, a longer list, or a PDF with the client's logo. The AI visibility checker is the three prompt sample when you do not yet have a list worth a weekly mail.
Related guides
- AI visibility for agencies when the memo is one of several client projects
- Share of voice in AI search for the figure in the share line
- How to check if Perplexity cites your brand when the action is a host, not a slogan
- Resources for the guide index
FAQ
How long should the weekly note be?
Short enough to forward. A header, separate sentences per engine, the movement, one wrong claim if any, and one action. If you cannot fill a line from the run, omit the line.
Which assistants can I name in the header?
Only the ones the run included. Paid plans can include ChatGPT, the second chat engine, and Perplexity. Free includes two of those. The upcoming engine is being added, so do not put the upcoming engine in the header until it is in the run. AI Overviews are a manual note, not a Mention Clock engine.
Do I need a cited source report to send the memo?
No. URLs are on each answer on every plan. The cited source report on Pro and Agency gathers them so you spend less time exporting. The memo still needs your judgment about which host matters.
Can one memo cover several clients?
No. One brand, one prompt list, one note. Mixing clients makes the share figure meaningless and makes a drop hard to assign. Agency workspaces exist so those projects stay apart.