What a monthly report is for
A monthly AI visibility report exists to answer one question: did anything change in what assistants say about us, and what should we do about it. If a page of the report does not help answer one half of that, it is decoration.
The report sits downstream of the measuring. Somebody, or some software, has asked a fixed set of questions in several assistants and recorded the results. If you are still working out how to do that part, start with our page on how to run the check. This page assumes you have the raw answers and want to know how to turn them into a document that people will read and act on.
There are two readers to keep in mind. One is the person who has to decide what to write or fix next, who needs detail. The other is the owner or client, who needs the finding in a minute and the evidence available if they want it. A good report serves both by putting the short version on top and the detail underneath.
The sections, in order
Seven sections, in this order. Not every business needs all seven in full, but each one should at least be considered.
1. The summary
Three to five sentences. What changed since last month, what did not, and the one thing to do. Written last, read first. If it can be pasted into an email unchanged, it is doing its job.
2. The setup note
A few lines: the number of questions, the assistants covered, the number of runs per question, the date range, and anything that changed in the method. This looks dull and it protects the whole report. When a figure moves, the first thing a skeptical reader asks is whether the test changed. The setup note answers before they ask.
3. The shares, by assistant
The percentage of answers in which you were named, cited with a link, and recommended first, shown per assistant, with this month beside last month and the number of answers behind each figure. Not a single blended score. If you want to see how the counts are built, read how to measure AI visibility.
4. The movement, with quotes
The questions where the outcome changed, in either direction, and the actual sentences. A rise from named to recommended, shown with the new wording, is more convincing than a percentage.
5. Who else is named
The businesses that appear when you do not, and any new arrivals. This is the competitive section, and it often surprises the reader.
6. Sources
Which sites the answers drew on this month, and whether your own pages were among them. Note any new source that began appearing.
7. What was done, and what is next
The work published or changed during the month, and the work planned for the next. This ties the numbers to effort, which is the only way to learn what moves them.
Writing the summary so it is honest
The summary is where reports quietly turn into marketing, so it is worth a few rules.
State movements in the same units as the body. If the body says named share went from twenty-two to twenty-six percent, the summary should say so, not call it strong growth. Let the reader decide whether four points is strong.
Say when a change is within normal variation. Because assistants give different answers on different runs, small moves are often noise. A line such as "within the range we would expect from run to run" is a sign of a report that respects its reader. Our page on why AI answers change explains how to judge what counts as normal.
Include one thing that did not work. A report that has only good news is not credible, and a reader who sees honest bad news will believe the good news more readily.
Finish with a single recommended action, not a list of twelve. If everything is a priority, nothing is.
What to ignore or cut
Reports in this field are padded with figures that look scientific. A few to skip, or at least to push to the back.
Blended scores with no breakdown. A single number from one to one hundred is easy to chart and impossible to act on. If the report offers one, it should show how it was built. Our page on the AI visibility score goes into what such scores hide.
Invented rank positions. Saying you were third in the answer treats a written paragraph like a search results page. The order of names in a sentence is not a rank you hold.
Wins from a single run. If a screenshot shows you recommended, ask how many runs it came from. One run is a coin flip.
Traffic as the headline. Most recommendations produce no click. A report that leads with sessions from assistants will undersell real progress. Traffic is fine as a supporting line, labeled as partial.
Sentiment scores with no quotes. A tone score cannot be checked. A quoted sentence can.
Dozens of charts. Each extra chart dilutes the three that matter. A report that fits on two pages and gets read beats a report of twenty that does not.
A layout you can copy
For a small business or a single client, the report can be a document of two to four pages. Here is a layout that works.
Page one holds the summary, the setup note and the shares table. The table has one row per assistant and columns for named, cited and recommended first, each showing last month, this month and the number of answers. A reader who stops after page one has the picture.
Page two holds the movement and the competitors. A short list of questions that changed outcome, with the old and new sentence, and a short list of who else was named and how often.
Page three holds the sources and the work. A list of the most cited domains and a note on whether your pages were among them, then the work done and the work planned.
An appendix holds the full answer log, so anyone who doubts a figure can read the original text. Our page on an AI visibility tracking spreadsheet shows how to keep that log.
Notice what is missing: no stock images, no definitions of what AI is, no pages of generic advice. Those add length and subtract trust.
How to judge a report someone sends you
If an agency, a consultant or a software product sends you a report, you can assess it in about five minutes with a handful of questions.
Does it say how many questions, which assistants and how many runs? If not, the figures have no foundation. Does each percentage show how many answers sit behind it? Is it split by assistant, or blended? Can you read the original answers? Does it quote the wording used about you? Does it list the businesses named instead of you? And does it say anything that did not go well?
Then ask whether the question set was the same as last month. If the supplier changed the questions, the month-on-month comparison means nothing, and the report should have said so loudly.
Finally, ask what you would do differently after reading it. A good report makes the next step obvious. A decorative one leaves you admiring the charts. For the step after reading, see our page on what to do after a check.
How often, and how much to read into one month
Monthly is the right rhythm for most small businesses. Weekly reports tempt people to react to noise. Quarterly reports are too slow to catch a mistake.
Do not read much into a single month. Because the same question can return different names on different runs, a small movement in either direction is weak evidence. Look for the same direction across two or three months, or for a large move that appears in more than one assistant at once.
Do read the quarter. Put three months side by side and look at the shape. That is where real progress, or a real problem, becomes visible. A quarterly review is also the right time to ask whether the question set still reflects how your buyers speak, and to add new questions as a separate series alongside the old ones.
The deeper audit, covered on our AI visibility audit page, belongs at the start and then perhaps once or twice a year. The monthly report is the light version in between.
Who should write it
Whoever does the work should not be the only person who checks the numbers. The temptation to round up is human, and it is stronger when the same person wrote the pages being measured.
If you are small, have someone outside the writing do the counting. A friend, a colleague, an assistant with a written rule sheet. If you hire a supplier, ask for the raw answers every month and read a few at random. You do not need to read all of them. You need to know that you could.
Whoever writes it, put the method note on page one and keep it identical from month to month. A stable method is the quiet foundation of every comparison in the document.
A sample summary, to show the tone
Here is what an honest summary can sound like. The numbers are invented to show the shape and are not a benchmark.
"This month we ran the same eighteen questions, four times each, in four assistants. Our named share was slightly up. Our recommended-first share did not change. Two questions about choosing between options moved from absent to named in one assistant, which we put down to the new comparison page published on the ninth, but it is one assistant and one month, so we are treating it as a hint. One question we had expected to win is still absent everywhere. The sources for it are two independent guides we are not in. Next month's job is to find out what those guides say about our category and whether we can be included."
Look at what that paragraph does. It gives the setup in one sentence. It reports the movement without cheering. It separates a hint from a result. It admits a miss. It names the next step. It takes under a minute to read, and the reader can go to the detail if they want it.
When the numbers are small
A small business testing twelve questions in two assistants has a small sample, and the report should say so plainly. Percentages from small samples jump around. Moving from two answers to three looks like a fifty percent gain and means almost nothing.
There are two sensible responses. The first is to show counts next to percentages, so the reader sees two out of twenty-four rather than eight percent. The second is to widen the sample before drawing conclusions, either by adding runs per question or by pooling two months of the same questions. Neither requires special software. Both make the report more honest.
It also helps to agree in advance what size of move you will treat as real. For example, you might decide that nothing under a certain number of answers counts as a change. The exact threshold matters less than having written it down before you saw the results, because a rule set afterward tends to bend toward the answer you hoped for.
Adjusting the report for different readers
The same data can serve several readers if you change the order and the depth, not the facts.
For an owner, lead with the summary and the one recommended action, and keep the rest short. For a content writer, lead with the questions where you are absent and the sources that are cited, because those decide what to write. For a client, lead with movement and the work done, since they want to know their money is producing something, and keep the setup note visible so the numbers are trusted. For a partner or investor, lead with the shares and the trend over three months.
Whatever the reader, never change the figures to suit them. Only the emphasis changes.
Reports from RunOctopus
For customers, RunOctopus includes reports on where ChatGPT, Perplexity, Claude and Gemini name the business. There is no free automated AI visibility checker, so these reports are part of the paid service, not a tool you can run on your own site for nothing.
The service itself is done for you. Ollie, the company's AI, researches the business, then writes and publishes pages directly on its own website, while a team plans, writes, reviews and publishes alongside it. Nothing is installed, approved, uploaded or scheduled by the owner. It works with Shopify, WordPress, WooCommerce and Wix. The 90-day promise is more visibility on Google and in AI answers within 90 days, measured against the 90 days before joining, or every dollar back.
To see what the plan for your site would look like, use the free site read at app.runoctopus.com/start. It reads your site and shows the plan. It does not test AI answers, and it needs no account and no call.