How to track your brand's visibility in AI search (AEO) with a spreadsheet

Many of your buyers no longer start with a list of blue links. They type "what's the best help desk for a small team?" into ChatGPT, Perplexity or Gemini, read one answer, and shortlist whatever it names. If your brand isn't in that answer, you lost the deal before your website had a chance, and nothing in your analytics will tell you it happened.

That is the problem Answer Engine Optimization (AEO) tries to solve. Before you can improve it, though, you need to measure it, and most teams have no idea how often AI assistants mention them, cite them or recommend a competitor instead. This guide walks through a simple, repeatable way to track that with a spreadsheet: which prompts to test, what to record, which numbers matter and how to turn the results into work your content team can actually do.

What AEO means, in plain terms

AEO (you will also see GEO, for Generative Engine Optimization) is the practice of making your brand show up accurately in answers written by AI systems: ChatGPT, Google Gemini, Claude, Perplexity, Microsoft Copilot and Google's AI Overviews at the top of search results.

Three things can happen to your brand in an AI answer, and they are worth separating from day one:

• Mentioned: the answer names you. This is the basic visibility signal.
• Cited: the answer links to your domain as a source. Citations send traffic and tell you which of your pages the engine trusts.
• Recommended: you are the first or top pick. Being mentioned fifth in a list of six is very different from being the answer.

A good tracking setup records all three, plus how the answer describes you, because an AI assistant that confidently gets your pricing wrong can do more damage than one that ignores you.

Why your normal SEO reports miss it

Rank trackers and Search Console were built for pages that rank in fixed positions. AI answers do not work that way:

• There is no position 1 to 10. An answer is a paragraph, a list or a comparison, and your brand is somewhere in it or nowhere.
• Answers change between runs. Ask the same question twice and you can get a different set of brands. One test tells you very little; a pattern over several weeks tells you a lot.
• Many answers produce no click at all. The buyer reads the recommendation and goes straight to the vendor's site or a review platform, so referral traffic from AI tools understates your real exposure.
• Each engine behaves differently. Some cite sources on nearly every answer, some answer from training data unless they search the web, and AI Overviews draw on pages from Google's own index. Strong in one engine does not mean strong in all of them.

So you need a different instrument: a fixed set of prompts, tested on a schedule, with every answer logged the same way.

The numbers worth tracking

Keep the scorecard small. These five metrics cover almost every question a marketing lead will ask.

Mention rate

The share of tested answers that name your brand. It is the headline visibility number, and the one to report weekly.

Citation rate

The share of answers that link to your own domain. Watch it alongside mention rate: being named but never cited usually means the engine knows you from third-party sites, not from your own content.

Share of voice

Your mentions divided by all brand mentions across you and your competitors. Mention rate tells you how visible you are; share of voice tells you whether you are winning or losing against the brands buyers compare you with.

Average position

When several brands appear, where do you land? Track the order in which brands are named, and count how often you are the top pick. Position is what separates "also ran" from "recommended".

Sentiment and accuracy

Record whether the answer is positive, neutral or negative about you, and whether its facts are right. A short "accuracy watch" list of wrong answers (wrong price, retired feature, missing free plan) gives your team a concrete list of pages to fix.

Build a prompt library that reflects real buyers

Your results are only as good as your prompts. Generic prompts produce generic results, so build the list from how real buyers talk. Good sources are sales call notes, support tickets, the questions in your demo requests and the queries in Search Console.

Aim for a mix of prompt types (the brand names below are fictional examples):

1. 01Category prompts: "What is the best help desk software for small businesses?"
2. 02Comparison prompts: "Compare Lumadesk vs Ticketora."
3. 03Alternative prompts: "Best alternatives to Ticketora for a growing support team."
4. 04Problem prompts: "How do I reduce first response time in customer support?"
5. 05Brand prompts: "Is Lumadesk worth it?" and "Lumadesk pricing."

Give each prompt a topic and a priority. Twenty to forty well-chosen prompts are plenty to start; it is better to test a small set consistently than a big one once.

How to run a test cycle

Consistency matters more than volume. A simple routine:

• Pick a cadence and keep it. Weekly or every two weeks works for most teams. Monthly is too slow to see whether a content change worked.
• Use the same prompts every time. If you need new ones, add them; do not reword old ones, or your trend lines stop meaning anything.
• Start a fresh chat for each prompt, with memory and personalisation off where the tool allows it, so earlier conversations do not colour the answer.
• Test every engine you care about. Twenty prompts across six engines is 120 answers per cycle. That sounds like a lot, but logging takes seconds per answer once your sheet is set up.
• Log one row per answer: date, engine, prompt, mentioned (yes/no), cited (yes/no), your position, sentiment, accuracy, which competitors appeared and the main third-party source cited.

Note which plan or mode you tested (free or paid, web search on or off), because answers can differ between them.

Using the AI Search & AEO Visibility Dashboard

If you would rather not build all of this from a blank workbook, the AI Search & AEO Visibility Dashboard is a ready-made Excel version of the method above. You still run the prompts and log the answers yourself; the workbook does all the scoring and reporting.

Here is how the tabs map to the routine:

• Prompts holds your prompt library with topic and priority, and gives each prompt a status from its test results: Winning, Contested, Gap or Invisible.
• Response Log is where each answer goes, one row per test, with room for 2,000 tests. Engine, competitor and sentiment fields are dropdowns, so the data stays clean.
• Dashboard shows your AI visibility score, mention rate, citation rate, share of voice, average position and top-pick answers, filtered by engine, topic and number of weeks, with the change against the previous period of the same length.
• Engines & Competitors scores each engine side by side and shows a brand-by-engine mention matrix for you and up to five competitors.
• Trends charts twelve weeks of every metric, per engine and per competitor.
• Sources counts which third-party sites the engines cite and flags high-priority sites where you are not listed yet, plus which of your own pages get cited.
• Action Plan links each optimization task to a prompt, so you can check whether its mention rate moved after you shipped the change.

The visibility score per answer combines mention, citation, position and sentiment, and the weights are editable on the Settings tab if your team cares more about citations than mentions, for example. Sample data from a fictional brand is included, so you can see every tab working before you replace it with your own.

Turning results into actions

The point of tracking is to know what to do next. The prompt status gives you a natural priority order:

• Invisible (no brand you track is named, or you never are): you probably lack content that answers the question directly. Publish a clear page that does.
• Gap (a competitor is named and you are not): study what the engine cites for that prompt. Often it is a comparison page, a review site or a community thread where the competitor is present and you are not.
• Contested (you and competitors share the answer): work on position. Clear positioning statements, honest comparison pages and up-to-date feature and pricing pages help engines describe you accurately.
• Winning: protect it. Keep the pages that get cited current, and watch the accuracy list for drift.

The Sources view tends to produce the quickest wins. If the engines keep citing the same review platform or forum for your category and you have no presence there, that is your outreach list. And every item on the accuracy watch list is a page on your own site to fix or clarify.

After you ship a change, keep testing the same prompts. A single good week proves nothing; a mention rate that climbs and holds over several cycles does.

Mistakes that make AEO tracking useless

• Testing once and drawing conclusions. Answers vary between runs. Trends over weeks are the signal.
• Rewording prompts mid-run. Add new prompts instead, so the history stays comparable.
• Tracking only one engine. Visibility in Perplexity says little about AI Overviews or Copilot.
• Counting mentions without position. Being listed last is not the same as being recommended.
• Ignoring wrong answers. Inaccurate answers about pricing or features cost deals quietly. Log them and fix the source.
• No competitor baseline. A rising mention rate means little if your main competitor is rising faster. Share of voice keeps you honest.

Common questions

How often should I test my prompts?

Weekly or every two weeks is a good rhythm for most teams. It is frequent enough to see the effect of a content change within a month or two, and light enough to keep up. Whatever you choose, keep the same prompts and the same routine every cycle.

Is AEO different from SEO?

It builds on SEO rather than replacing it. Pages that rank well and answer questions clearly are often the ones AI systems cite, especially in AI Overviews. What changes is the measurement: you track mentions, citations and share of voice in answers instead of positions on a results page.

Can I do this in Google Sheets?

You can import the workbook into Google Sheets, but it is built for Microsoft Excel (desktop or Excel for the web). Sheet protection does not carry over to Google Sheets, so the locked formulas are easier to overwrite there by accident. If your team works in Excel, use it there.

Do I need paid AI subscriptions to track this?

Not necessarily, but be consistent. Free and paid tiers, and modes with web search on or off, can give different answers to the same prompt. Decide which version of each engine reflects your buyers, note it in your log and test the same way every time.

Tracking AI visibility does not need an expensive platform to get started. A fixed prompt library, a disciplined testing routine and a workbook that does the maths will show you where you stand within a few weeks. You can preview the AEO Visibility Dashboard free; downloading it is included in the OpenWorksheet monthly plan, along with the rest of the library.