Why a plain spreadsheet is enough
A plain spreadsheet is the right tool for tracking AI visibility by hand. It holds your fixed questions, a column per assistant, and a record of who got named. It costs nothing, it lives in an app you already have, and it makes the month-to-month comparison easy, which is the entire reason to keep records.
You do not need a template from the internet, and you should be wary of any that promises a score at the bottom. A score computed from three weak inputs is a weak score. What you need is a clear place to write down what you saw, in the same way every month.
This page describes the layout in words so you can build it in any sheet app in about ten minutes. If you are not yet running checks, start with the free manual check and come back. And if you want the wider picture of checking, our broader guide to checking sets out what a check can and cannot show.
One tab for questions, one tab for answers, one tab for a monthly summary. Same columns every month. Never edit old rows.
Three tabs: questions, answers, summary
Keep the sheet in three parts. Separate tabs make it hard to damage your own history.
- Tab one, Questions. The frozen list of questions. One row each. This tab changes only by adding rows at the bottom.
- Tab two, Answers. One row per question per assistant per run. This is where you write what you saw, and it grows every month.
- Tab three, Summary. A few counts, drawn from the Answers tab. This is what you read.
The separation matters for a practical reason. The Answers tab gets long. A set of twenty questions across four assistants is eighty rows a month, nearly a thousand rows a year. Nobody wants to scroll that to find out how things are going, which is why the Summary tab exists.
If you prefer one tab, you can, but keep the question list at the top where it will not be edited by accident. A simple rule: if a cell holds something that should never change, put it somewhere you will not be typing.
The Questions tab
Four columns are enough.
- ID. A short number, Q01, Q02 and so on. You will use it in the Answers tab.
- Question text. The exact wording you ask. Do not edit it after the first run.
- Kind. Shortlist, problem, comparison, constraint, place or doubt. This lets you see later whether you are strong at one kind and weak at another.
- Date added. So you know which questions have the longest history.
How to write the questions themselves, and why the wording must stay fixed, is the subject of how to write the questions you test with. This page assumes you already have your set, or will have it before the first run.
If you want to retire a question, do not delete the row. Add a column called Status, and mark it "retired" with the date. Old answers stay in the Answers tab, still linked to a question that still exists. That keeps the history clean.
The Answers tab, column by column
This is the heart of the sheet. One row each time you ask one question of one assistant. Use these columns.
- Run date. The day you asked.
- Question ID. Q01 and so on, matching the Questions tab.
- Assistant. ChatGPT, Perplexity, Claude, Gemini, or Google AI Mode if you check it.
- Logged out? Yes or no. If no, say why in the notes.
- Were you named? Yes or no.
- Position. First, in a list of several, or passing mention. Leave blank if not named.
- Linked? Whether one of your pages was shown as a source, and which.
- Others named. Every other business, in order, separated by commas.
- Sources shown. The site names of the pages the assistant listed or drew on.
- Your wording. The exact sentence about you, if you were named.
- Notes. Anything odd: a wrong fact, a refusal, a sign-in prompt.
That is eleven columns, and it sounds like a lot. It is quick in practice, because most cells are one word. The ones that take thought are "Others named" and "Sources shown", and they are the two that make the sheet useful. They feed the work described in checking competitors in AI search.
The Summary tab: four counts
Keep the Summary short. Four counts are plenty, and each one is a simple count of rows in the Answers tab for a given run date.
- Named in X of Y answers. Total rows where "Were you named" is yes, over total rows for that run.
- Named first in X answers. Rows where Position is first.
- Linked in X answers. Rows where one of your pages was shown.
- Per assistant. The first count, split by assistant.
Add one more block if you can, a list of the three or four other businesses named most often, with their counts. It is the most revealing part of the sheet, and it takes a minute to make with a simple count by name.
Resist inventing a combined score. Our page on the AI visibility score explains why one number can hide more than it shows. "Named in 14 of 80 answers" is clear, honest and comparable. "Visibility 17.5" is a number that needs a footnote.
If you want to go further on what counts and how to weigh it, the page on how to measure AI visibility covers it.
The monthly routine
A repeatable routine matters more than a clever sheet. Here is one that takes about an hour for twenty questions.
- Put it in your calendar for the same week each month.
- Open a private window and the assistants you track.
- Add the run date to a fresh block of rows in the Answers tab, one row per question and assistant, so the blank rows are waiting for you.
- Ask each question once, exactly as written, and fill in the row as you go. Do not wait until the end.
- Check the Summary and write two sentences underneath. What moved, and what did not.
- Save a copy of the sheet with the date in the name, as a backup.
The diagram above shows what you gain over time. The first month is only a baseline. The first comparison, a month later, is mostly noise, because assistants vary from one run to the next. The third run is the first point where a trend starts to mean something, and the sixth month is where you can start asking whether your own changes moved anything.
That is slow, and it is meant to be. If you want a quicker read on whether a single answer is steady, see why AI answers change.
Habits that keep the data honest
Several small habits make the difference between a sheet that helps and one that misleads.
Never overwrite. Add rows. Do not edit a past month, even to fix a mistake. If you made an error, add a note in a Notes column.
Write the observation, not the interpretation. "Named third, in a list of five" is an observation. "Doing well" is a feeling. Record the first. Keep the second for the end of the month.
Copy the wording. The exact sentence is worth more than your paraphrase. It lets you notice a change in how you are described, long before a change in whether you are named.
Record failures too. If an assistant refused to answer, asked you to sign in, or gave an empty reply, write that down. A gap in the data is data.
Keep your own bias out of the cells. If you notice yourself reading answers charitably, have a second person fill in a sample of rows and compare.
Back it up. Keep one copy somewhere other than your own computer, in case the file is lost. The record is the thing you cannot recreate.
Reading the trend without fooling yourself
When the third or fourth month is in, you can look at the numbers and ask whether anything is real. A few rules keep you honest.
Small moves are noise. If you were named in 14 of 80 answers last month and 16 of 80 this month, that is not a change. Assistants vary. Look for moves that persist across runs.
Look at per-assistant counts. A rise in one assistant and a fall in another may cancel out in the total, and hide something useful.
Look at the competitor block. A change in who else gets named is often more telling than a change in your own count.
Look at the wording column. A shift from "also worth a look" to "a strong choice for" is a real change that a count will not show.
Match changes to actions. Keep a short log in the sheet of what you did and when: published a guide, added reviews, updated a listing. Without it, a good month has no explanation.
For how to turn results into actions, see what to do after an AI visibility check. For how a more formal report is structured, see what a monthly AI visibility report contains.
When a spreadsheet stops being enough
A spreadsheet works while the set is small and you keep the habit. It starts to strain in three situations.
The set grows past about thirty questions. Entering a hundred and twenty rows a month by hand is a lot, and people stop.
You need history at a pace faster than monthly. If you want weekly or daily readings, hand entry will not hold.
More than one person must run it. Without discipline, different people record differently, and the data becomes inconsistent.
At that point, you either shrink the set, hand the task to someone who will do it consistently, or move to software or a service. Our page on AI visibility monitoring compares one-off checking with continuous watching, and AI visibility tools describes the categories of software on the market.
One honest suggestion: do the spreadsheet for at least three months before deciding. You will learn what you actually want to measure, and what to ask of any tool or service later. A person who has run it by hand knows which questions matter.
A sheet with three tabs, eleven columns and a fixed question list is enough for a small business to track AI visibility honestly. Add rows, never edit them.
A worked example of three filled rows
It helps to see what real rows look like. Here are three, described in words, for an imaginary pet groomer in a mid-sized town. The details are invented, only the shape matters.
Row one. Run date is the first of the month. Question ID is Q03, a place and job question about mobile dog grooming. Assistant is Perplexity. Logged out is yes. Named is no. Position is blank. Linked is no. Others named lists three groomers in a comma-separated string. Sources shown lists a local directory, a regional news site and a forum. Wording is blank. Notes says the answer led with a general tip before naming anyone.
Row two. Same date, same question, assistant is ChatGPT. Named is yes, position is third in a list of four, linked is no. Others named lists the same three plus the groomer. Wording quotes the sentence, which calls the business "good for anxious dogs." Notes records that no sources were shown.
Row three. Same date, Q03 again, assistant is Gemini. Named is no. Others named lists two businesses. Sources lists a map listing site. Notes says the answer asked for a location.
On the Summary tab, Q03 shows named in one of three assistants checked, and the competitor block shows one groomer appearing in all three answers. That groomer is the one to study.
Three rows of work, five minutes each, and already something to act on.
What RunOctopus can and cannot do for the sheet
There is no spreadsheet or free checker from RunOctopus that fills this in for you. The sheet in this guide is yours to build and keep.
RunOctopus is a done-for-you service for the work a sheet like this points at. Ollie, our AI, researches your business and writes and publishes pages on your own website, with a team that plans, writes, reviews and publishes alongside it. It works with Shopify, WordPress, WooCommerce and Wix, and you do not install anything or approve each page. Paying customers receive reports on where ChatGPT, Perplexity, Claude and Gemini name their business, so they do not have to keep a sheet by hand.
If you are keeping a sheet, you may still want your own record alongside any report, and that is fine. The sheet and a report answer the same question from two sides.
To see what Ollie would plan for your site, try the free site read. You paste your address, and it reads the site and shows the plan. No account and no call. It does not check AI answers.