# Build a Free AI Mention Tracking Tool in a Spreadsheet

> A spreadsheet template, a prompt list and a weekly routine that work as an AI mention tracking tool for your SaaS, using only free accounts and no code.

Source: https://comppound.com/blog/build-a-free-ai-mention-tracking-tool-in-a-spreadsheet (published 2026-11-07)

You can build an AI mention tracking tool from a spreadsheet, a fixed list of buyer questions, and a weekly routine of asking those questions and logging what the AI assistants say. Record whether your brand is named, whether your URL is cited, and which competitors appear instead, then read the trend rather than any single answer.

## What you will have at the end, and what you need first

At the end you will have a Google Sheets or Excel file with three tabs: a prompt list, a run log, and a summary that shows how often each assistant names you. It takes an afternoon to set up and about half an hour a week to maintain.

You need a free account on each AI assistant your buyers are likely to use, a spreadsheet, access to your analytics, and a list of your five closest competitors. The steps, in order:

1. Write 15-20 prompts that mirror how buyers ask.
2. Build the log tab and the summary tab.
3. Run every prompt on a schedule, in clean conditions.
4. Score each answer on the same fields every time.
5. Read the summary against your analytics and decide what to change.

Throughout, take a company of 40 people that sells approval-workflow software to finance teams. It has a blog, one in-house writer, and no idea whether any assistant ever names it. Everything below is applied to that company.

## Write the prompt list before you open any assistant

The prompt list is the part of the system that decides whether the data means anything. Write prompts the way a buyer types them, not the way you would write a keyword. For the approval-workflow company, "best invoice approval software" is a keyword. "We're a 30-person finance team and invoices get stuck in email, what should we use?" is a prompt.

Split the list into four groups: category questions that name no brand, comparison questions that name a competitor, problem questions that describe the pain without naming a product category, and a few branded questions that name you directly. Keep the branded group small. Those prompts almost always mention you, so they tell you about accuracy, not about visibility. For more on finding the phrasings that actually surface a category, see [testing which AI prompts mention your SaaS category](/blog/testing-which-ai-prompts-mention-your-saas-category).

Give each prompt an ID and freeze the wording. Changing a prompt midway breaks the comparison with every earlier row, so if a prompt turns out to be bad, retire it and add a new ID instead of editing it.

You will know this step worked when someone outside marketing, ideally a sales rep, reads the list and says buyers really do ask these things.

## Build the log tab and the summary tab

Create the log tab with one row per answer, not one row per prompt. This matters because the same prompt can produce different answers each time, and you want every run on record.

| Column | What goes in it |
|---|---|
| Date | The day you ran it |
| Prompt ID | The ID from your prompt tab |
| Assistant | Which assistant produced the answer |
| Run | 1, 2 or 3 for repeats of the same prompt that day |
| Named | Y or N: is your brand named anywhere in the answer |
| Position | Where you appear if the answer is a list, blank if not |
| URL cited | Y or N, plus the exact URL if yes |
| Competitors named | Names, separated by commas |
| Source domains | Any sites the assistant lists as sources |
| Notes | Anything wrong, outdated or oddly phrased about you |

On the summary tab, use COUNTIFS to turn the log into rates. For one assistant, divide the count of rows where Assistant matches and Named is Y by the count of rows where Assistant matches. Build the same cell for URL cited, and for each competitor. A pivot table does the same job if you prefer it.

You will know it worked when you can type a few fake rows into the log and see the summary move.

## Run each prompt several times, in clean conditions

![Summary tab showing the share of runs per AI assistant where a SaaS brand is named versus where its URL is cited](https://images.unsplash.com/photo-1648134859179-ed0c98f54519?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3wxMDMwMjQ2fDB8MXxzZWFyY2h8Mnx8QUklMjBtZW50aW9uJTIwdHJhY2tpbmclMjB0b29sfGVufDB8MHx8fDE3OTQwODA3MzV8MA&ixlib=rb-4.1.0&q=80&w=1080)
*Photo by [Team Nocoloco](https://unsplash.com/@teamnocoloco?utm_source=comppound&utm_medium=referral) on [Unsplash](https://unsplash.com/?utm_source=comppound&utm_medium=referral)*


Answers from an AI assistant are not fixed. Ask the same question twice and the brands named can differ, so one run tells you almost nothing. Run each prompt three times per assistant, in a fresh conversation each time.

Reduce personalisation where you can. Use a logged-out session or a chat with memory features turned off, and do not run prompts in a conversation where you have already discussed your company. If you test from a company account that has talked about your product before, the assistant may lean on that history and flatter your numbers.

Pick one day and one time block each week and keep it. Twenty prompts across three runs and a few assistants is a lot of typing, so cap the total at what fits in thirty minutes. If it does not fit, cut prompts, not runs. A shorter list you actually finish every week beats a long list you abandon in week three. The [manual AI brand visibility audit](/blog/ai-brand-visibility-audit-manually-check-if-answers-cite-your-saas) walks through the same routine at a smaller scale if you want to start there.

## Score every answer with the same rules

Decide scoring rules once and write them in a note at the top of the sheet. "Named" means the exact brand name appears in the answer text. A vague reference to "tools like this" does not count. "URL cited" means a link to your domain appears as a source or in the text, and it only counts after you click it and confirm it loads.

That click matters. Assistants sometimes show a link that does not resolve or that points to a page that never said what the answer claims. Mark those in Notes, because they are a different problem from not being cited at all.

For the approval-workflow company, the first week might show the brand named in a few comparison answers, never in the category answers, and a competitor appearing in nearly every category answer. That pattern points at a specific gap: the site has no page that makes the category case. It is more useful than a general feeling of being invisible.

Also record what the answer gets wrong. If an assistant describes your pricing model incorrectly or lists a feature you dropped, put it in Notes. Those rows become the content fixes, and the guide on [how to write an answer AI can quote](/guides/ai-search/how-to-write-an-answer-ai-can-quote) covers how to correct them on the page.

## Read the results against your analytics, then change one thing

After four weeks of data you have a baseline. Before that, the summary is noise, so resist reacting to week one.

Look at three things. First, the gap between "named" and "URL cited": being named without a link means the assistant knows you from somewhere else, which is a clue about where your reputation lives. Second, the source domains column: if the same few third-party sites keep appearing as sources, those are the places where being listed or reviewed matters most. Third, which competitors appear when you do not.

Then check your analytics for visits that arrive from AI assistant domains. Look in your referral or traffic-source report and filter by the domains of the assistants you track. Treat this as a partial view, because not every click from an assistant carries a referrer. The methods in [AI conversion tracking for B2B SaaS content](/blog/ai-conversion-tracking-methods-for-b2b-saas-content) cover how to tie those visits to signups.

Change one thing at a time, such as publishing a comparison page or updating an old post, and note the date in the sheet. If you change five things at once you will never know which one moved the numbers. Comppound runs this kind of prompt tracking inside its GEO work, with one person reviewing each account's results, because someone has to decide what a shift in the data means.

## What goes wrong, and how to recognise it

**Your mention rate swings wildly from week to week.** The symptom is Y in one run and N in the next for the same prompt. This is normal variation, not a broken sheet. Judge by the rate across runs and across weeks, and do not treat a single flip as a result.

**Your numbers look great but nobody is buying.** The symptom is a near-perfect mention rate. Check how many of your prompts name your brand or your exact product category. Branded prompts inflate the score, so report the unbranded group separately.

**The sheet goes quiet.** The symptom is gaps in the Date column. The routine is too heavy. Cut the prompt list until it takes half an hour, and put the session on a calendar as a recurring slot.

**Competitor counts look too low.** The symptom is rows where a competitor was clearly named but not logged. People misspell and abbreviate names. Keep a list of name variants at the top of the sheet and score against it.

**The trend line breaks for no reason.** The symptom is a sudden jump on the day someone else took over the logging. Different people score "named" and "cited" differently. Write the scoring rules down, and have both people score the same ten answers to check they agree.

**You outgrow the sheet.** The symptom is that logging takes more than an hour a week, or you need many more prompts and assistants. At that point compare options honestly, using [citation platform versus custom tool](/blog/citation-platform-vs-custom-tool-for-a-60-person-saas) and [building automated AI citation tracking](/blog/how-to-build-automated-ai-citation-tracking-saas-for-your-blog). Automation can collect the answers, but a person still has to judge whether a change in the data deserves a change in the roadmap.

[See how Comppound approaches SEO for B2B SaaS](/services/seo-growth).

## Frequently asked questions

### How often should I check AI mentions and citations?

Weekly is enough for most SaaS teams. Answers shift run to run, so daily checks mostly record noise, while monthly checks leave too few data points to see a trend. Keep the day and time block the same each week so the rows are comparable. Review the summary monthly and act on patterns that hold across several weeks.

### Does a mention without a link still count?

Yes, but track it separately from a cited URL. A mention without a link shows the assistant associates your brand with the topic, though it sends no one to your site. Compare how often you are named with how often you are linked. A large gap suggests your reputation lives on other sites, and your own pages are not the ones being used as sources.

### Can I automate this spreadsheet with a script or an AI agent?

You can automate collecting the answers, and for a long prompt list that is worth considering. Scoring is harder to hand off, because deciding whether an answer is accurate or just names you in passing takes judgment. If you automate, sample a handful of answers by hand each week to confirm the automated scoring still matches what you would have recorded.

### Why does the same prompt give different answers in different sessions?

These assistants generate each answer fresh, and some also draw on conversation history, location or account settings. That is why the method here repeats each prompt and uses clean sessions. If you see an answer that flatters you once, treat it as one sample, not a result. Only a rate across many runs is worth reporting to anyone else.

## About Comppound

Comppound runs SEO and AI search for B2B SaaS companies: a person sets the plan and reviews samples and every high-risk page, and AI speeds up production.

- [See SEO Growth](https://comppound.com/services/seo-growth): Buyers compare before they book a demo. Are your pages there when they do?
- [See pricing](https://comppound.com/pricing): Want the content written, reviewed and published for you?