A monthly GEO report template for agencies

A monthly GEO report for agencies is a report that shows, in one document, how a client brand appeared in AI answers that month, which questions and engines it was measured with, and what could not be measured. A good template has seven sections: scope, results as count pairs, what could not be measured, the sources shown, accuracy findings, period comparison and actions. Below you will find what each section contains and an example table; the numbers in the example tables do not come from a real measurement, they only show the format.

The report's skeleton

We wrote the template section by section, in a form you can copy and fill in every month:

1. Scope: Which questions, which engines, which dates.
2. Results: Separately for each engine, as count pairs.
3. Unmeasured answers: Answers that returned an error, were cut off or could not be read.
4. Sources shown: The sites the engines relied on to build their answers.
5. Accuracy findings: Wrong or incomplete information the engines stated about the brand.
6. Period comparison: Only if it was done with the same questions and the same engines.
7. Actions and the next measurement: What will be done, which finding it rests on, and when it will be measured again.

There is a logic to this order. Readers should see what a number covers before they see the number, and right after the results they should read what could not be measured. If the unmeasured answers are pushed to the end of the report, the results look more complete than they are.

1. Scope: questions, engines, dates

The scope section is the report's first screen. It states three things: the question set, the engines and the measurement dates. The question set should have a name or a version number; that is how the reader can tell the same set was used the following month.

Google's two AI surfaces should also be treated separately. According to the Google Turkey Blog, AI Mode and AI Overviews began rolling out gradually in Turkey from 18 February 2026. Google's documentation also says the two may use different models and techniques, and that the answers and links they show will differ. If you also measure AI Mode, put it on a separate row; if you do not, mark it as "not measured" in the scope table. Merging it with AI Overviews hides the difference between them.

Questions that name the brand are kept separate. In a question like "Is brand X trustworthy?", X's name already appears; such a question shows not visibility but what the engine knows about the brand. Adding both kinds of question into a single rate makes the rate look higher than it is.

2. Results: each engine separately, as a count pair

Results are written not as a single percentage but as a count pair: "9 of 38 answers". If a percentage is wanted, it goes next to the pair, not in its place. On its own, "0%" may not be a finding; "0 of 116 answers" is a finding. We published our own product's measurement of 27 August 2026 in exactly this form on the What is GEO? page: 26 questions, 5 engines, and our product was not mentioned in any of the 116 measurable answers.

Example table (the numbers do not come from a real measurement):

The Google AI Overviews row needs a separate detail. Google does not show an AI Overview for every search. A search with no AI Overview is a "measured, absent" state and is real information; an error from the service that fetches the search results is a "not measured" state. If both are written as the same zero, an outage turns into a "Google does not show you" finding.

You can also add a "how it was mentioned" column to this section: as the leading option, as one of the alternatives, or with a caveat. Do not merge the engines into a single composite score; we explained why they cannot be merged in our post on why the same Turkish question returns different brands.

3. Unmeasured answers are written separately

This section determines whether the report is honest. An engine can time out, a provider can return an error, a source list can fail to come back. For these answers you do not have the information "not mentioned"; you have no information.

We know how often this happens from our own measurement. In our measurement of three corporate brands between 16 and 25 August 2026, 189 of 1,835 planned answers (10.3%) could not be measured: 166 returned an error, 16 came back without source information, and 7 had no AI Overview block to read. These answers were kept out of the rates. This measurement is limited to three brands; the rate may differ for another client, but you should assume it will not be zero.

Example table (the numbers do not come from a real measurement):

Even if this section is empty, it should stay in the report with its heading: writing "No answers went unmeasured this month" is different from leaving the section out altogether.

Next step

Let us plan a measurement for your brand

The quote page explains what a paid measurement covers and what sets its price.

Go to the quote page

4. Sources shown

This section shows which sites the engines relied on to build their answers. The list of sites is grouped by domain, and next to each site it says which engines showed it. The client's own site sits on a separate row.

Example table (the numbers do not come from a real measurement):

Do not forget to state the base. In this example, the base is the 61 answers that cited at least one source. Some engines show no source at all in a large share of their answers; taking all answers as the base makes those engines look different from what they are. The base used should be written above the table.

Being cited as a source and being mentioned are different things: an engine can read the client's site and not say its name. We explained this distinction with our own data in our post on the difference between a brand mention and a source citation.

5. Accuracy findings

This section states whether what the engines said about the brand is correct: founding year, address, product range, cities served and the like. Each finding is given together with the answer it came from, the engine and the date; the client should be able to read the sentence with their own eyes.

The accuracy section writes three states separately:

This distinction is not a small detail. For a client whose site is closed to crawlers, there is often no information to compare the engines' statements against; and this client may also be the one most at risk of wrong information. If "no wrong information found" and "could not be checked" are written on the same row, the report tells the riskiest client the most reassuring sentence. For how to find and correct wrong information, see our post on detecting and correcting false information in AI.

6. Period comparison: the same-questions condition

Two months are compared only if they were measured with the same question set and the same engines. If the question set changed, the difference between them may come from the changed questions rather than from the agency's work. In that case the report does not write an increase or a decrease; it writes "not comparable because the question set changed".

Even two months measured with the same questions call for care with small differences. AI answers can change for the same question: in research published by SparkToro in January 2026, in trials where 600 volunteers asked 12 questions a total of 2,961 times, the chance of the same brand list coming back in two answers was under one percent on ChatGPT and on Google's AI. So a move from "9 of 40 questions" to "10 of 40 questions" from one month to the next may not be a real change.

Example table (the numbers do not come from a real measurement):

Compare how often you appear rather than your rank. The sentence "We moved from third to second on ChatGPT" may no longer hold in the next answer; how many answers you were mentioned in for the same questions is a sounder comparison.

7. Actions and the next measurement

The last section turns the report into a work list. Next to each action it states which finding it rests on, for example: "On these three questions the engines cited a complaints site and did not mention the brand." An action with no finding behind it does not go into this section.

Actions are not written with a promised outcome. A sentence that says "If you do this, you will be mentioned this much more" has no measured evidence behind it. Instead, each action states how it will be checked in the next measurement: which questions, which engines.

The section closes with the date of the next measurement and a note that the same question set will be used. That way, the next report's period comparison is possible from the start.

How does GeoMono cover this template?

You can fill in the template with any tool, or even by hand. For agencies using GeoMono, Reports and Agency Panel does the following today: client brands are managed from one account, and each brand has its own workspace, question set and report. The report is shared through a link the client can open without signing in; the person who receives the link sees the report and the evidence pages, but cannot enter the workspace.

The period report compares two measurements only if they were made with the same questions; if the difference is at noise level, it says "no change detected". Steps are sorted into now, next, investigate and monitor bands. The period report is published with the agency's name and logo and can be printed to PDF from the browser.

Here is what does not exist yet: there is no separate PDF export and no public API. The agency's name and logo currently appear on the period report; the panel and the other report pages carry the GeoMono name.

Frequently asked questions

Which sections should a monthly GEO report have?

Seven sections are enough: scope, results engine by engine, unmeasured answers, sources shown, accuracy findings, period comparison and actions. Scope comes first, because readers should know which questions and engines a number was measured with before they see it. Unmeasured answers sit right after the results.

Why should we write results as count pairs instead of percentages?

A percentage hides how many answers it was calculated from. "0 of 116 answers" and a bare "0%" do not give the same information: the first is a measured result, while the second could also be a number in which unmeasured answers were mixed into the zero. If a percentage is wanted, it goes next to the count pair.

If we change the question set, can we compare with the previous month?

You cannot. If the question set changed, the difference between the two months may come from the changed questions rather than from the work done. In that case the report does not write an increase or a decrease; it records the new set as a baseline and compares against it the following month.

Should unmeasured answers be shown to the client?

They should. In our own measurement, 189 of 1,835 answers planned between 16 and 25 August 2026 (10.3%) could not be measured. Had these answers been counted as zero, the client would have read "you were not mentioned" for answers that were never measured.