GEO metrics for your marketing dashboard: what to report every month

GEO metrics measure how AI assistants treat your brand: whether they name you (presence), whether they name you first (first position), how much of the conversation you hold (share of voice), whether they cite your pages (citation share) and whether what they say about you is true (description accuracy). Together they answer the question a marketing director actually asks: when buyers ask AI about our category, do we come up, and is what it says right?

These five belong on the same dashboard as your search and paid media numbers. They need different handling, though, because AI answers vary between runs and engines in a way rankings do not.

In short

• Report five metrics: presence, first position, share of voice, citation share and description accuracy.
• Report every metric per engine, with the number of answers behind it.
• Show no percentage resting on fewer than 30 answers.
• Keep branded and unbranded questions apart. Unbranded is the headline.
• Treat clicks from AI features as a supporting number, not the main one.

Which GEO metrics should you report?

Metric	Question it answers	How to calculate it	Watch out for
Presence	Are we in the answer at all?	Answers that name your brand ÷ all answers	Branded questions inflate it; report them separately
First position	Are we the first brand named?	Answers that name you first ÷ all answers	An answer that names one brand makes it first by default
Share of voice	How much of the conversation do we hold?	Your brand mentions ÷ all brand mentions	Changes if the competitor list changes; keep it fixed
Citation share	Do answers use our pages as evidence?	Citations of your domains ÷ all citations	Answers written from memory cite nothing
Description accuracy	Is what AI says about us true?	Answers that describe you correctly ÷ answers that describe you	Needs a written fact sheet and a human reviewer

Every metric is calculated per engine and per question group, on a fixed question set. Our guide to how to measure GEO covers the sampling behind them.

Presence and first position

Presence is the base metric. If an assistant does not name you, nothing else about the answer helps you. Report it on unbranded questions as the headline, because a question that names your brand will almost always mention you.

First position matters because buyers read from the top. Researchers count it too: the paper that coined the term GEO (Aggarwal et al., KDD 2024) measured visibility as a position-adjusted word count, weighting each source by where in the answer its text appeared. Read first position next to presence. With 15% first position and 40% presence, you lead in about three of every eight answers you appear in, and the job is to move up. With 15% first position and 16% presence, you lead almost every time you appear, and the job is to appear more often.

Share of voice

Share of voice compares you with your competitors. Count each brand once per answer, then divide your mentions by the total. It is the metric to put next to market share, and it only works if the list of competitors you count is fixed and written down. Our explainer on AI share of voice goes deeper on how to set it up.

Citation share

When an assistant searches, it cites sources. Citation share is the part of those sources that are your own pages. Expect it to look small. In our analysis of 96 Google South Africa searches on 11 October 2026, an AI Overview cited 10.9 sources on average, and the Overviews together cited 412 different domains. Six of the 92 Overviews listed no sources at all.

Track two figures, then: your share of all citations, and the share of answers that cite your site at least once. Remember that an answer can name you without citing you, and cite your page without naming you. Both are worth knowing.

Description accuracy and sentiment

Accuracy is the metric most dashboards miss, and the one most likely to cost a sale. Write a one-page fact sheet: what you sell, where you operate, published prices or fees, opening hours, how to contact you. A reviewer then marks every answer that describes you as correct, out of date, wrong or confused with another business. Errors point to a fix: an old page, an outdated listing, a similarly named company. Our guide on how to stop AI confusing your brand covers the last one.

Sentiment (positive, neutral or negative) is useful but subjective. Report it, but lead with accuracy: you can act on a wrong fact this week.

A sample monthly GEO report

The report below is illustrative. Meridian Bank is a fictional brand and every number is invented to show the layout. It assumes 120 unbranded questions, asked once a month on each of three engines: 360 answers a month, 120 per engine.

Metric (unbranded questions)	September	October	Change	Base in October	How to read it
Presence, all engines	38.1%	41.4%	+3.3	360 answers	Within normal variation
Presence, engine A	45.0%	52.5%	+7.5	120 answers	Promising; confirm next month
Presence, engine B	40.0%	41.7%	+1.7	120 answers	Flat
Presence, engine C	29.2%	30.0%	+0.8	120 answers	Flat, and the weakest engine
First position, all engines	14.2%	15.0%	+0.8	360 answers	Flat
Share of voice, all engines	17.6%	18.9%	+1.3	788 brand mentions	Flat
Citation share, all engines	2.1%	3.4%	+1.3	2,960 citations	Up; check which page is newly cited
Description accuracy	88.3%	91.3%	+3.0	149 answers naming Meridian	Two old fee figures still quoted
Presence, business banking, engine C	Not enough answers	Not enough answers	None	24 answers	Below the 30-answer floor

Two rules make this table honest. Every figure carries its base, and a change is read against its margin of error. On 360 answers a month, a move near 40% needs about 7 points to be clearly real, so +3.3 is noise. On 120 answers per engine the bar is about 12 points, so engine A's +7.5 is a lead to follow, not a result. Where the base drops below 30, the table says so instead of printing a number.

What about traffic and conversions?

Keep them on the dashboard, beside the answer metrics. Google says traffic from its AI features is reported in Search Console's Performance report, within the "Web" search type. Clicks matter, but they only count the buyers who click. An answer that names you, describes you correctly and puts you first can do its work without any click at all.

Two cheap supporting signals help close the gap: branded search demand over time, and a "where did you hear about us?" question on your enquiry form with "an AI assistant" as one of the options.