GEO for agencies: tracking your clients’ AI visibility

Choosing a client’s prompts, organising the tracking, reporting every month: a method for agencies.

Christophe Tran Founder of iPand Oct 23, 2026 · 6 min

What should you track for a client?

For each client, track three things: whether they are cited in AI answers on their key questions, who gets cited instead, and which sources those answers rely on. Everything else follows from those three measures.

The client asks a simpler question: “are we in ChatGPT?”. The honest answer is a share of answers, not a position: SparkToro showed that the list of recommended brands changes almost every time the same question is asked again, to the point that a “ranking in AI” makes no sense^[1]. That is the first thing to explain to the client, before the first number.

Depending on the client, add the tone used about them and the technical readability of their pages. Our article on six metrics covers each one.

Choosing a client’s prompts

Start from the questions your client’s customers actually ask, not from their keyword list. A prompt is a full question with its context: a city, a budget, a use case.

Three sources are enough to find them: the questions that reach sales and support, Search Console queries, which group data by the query users typed^[2], and the questions assistants already raise when you ask them about the industry. Sort them by intent: discover, compare, choose, buy.

Keep branded prompts (“is [brand] any good?”) and unbranded ones (“which firm should I hire for…?”): the unbranded ones are what really measure visibility.
Cover each tracked market in its own language: the same question asked in the US and in the UK does not always get the same answers.
Freeze the list for a period, then review it on a set date: changing prompts every week makes the curves unreadable.

Volume depends on the client’s ambition. In iPand, a plan tracks from 50 to 350 prompts depending on the tier^[3]: enough to cover a small business’s main intents, or several product lines and markets for a larger brand.

Organising the tracking: one brand, one workspace

Each client gets its own tracking space, with its own prompts, competitors and history: mixing two clients in one setup muddles every measure. That is how most tools work, too.

In iPand, one subscription tracks one brand: each client takes the plan that suits it, and subscriptions are combined in the same agency account, with a brand selector^[4]. Team members are unlimited, so you can invite the client to look at their own results^[3]. One subscription therefore does not cover several clients: each client brand has its own.

Whatever the tool, document each client’s prompts, competitors and markets, and the date of every action you take: that is what makes the report credible. Our comparison of tools covers the other options on the market.

Your agency account

Client brand A Its subscription, prompts and competitors
Client brand B Its subscription, prompts and competitors
Client brand C Its subscription, prompts and competitors

Subscriptions add up in the same account, with a brand switcher.

One subscription tracks one brand: each client has its own tracking.

Reporting to the client

A good monthly report fits on one page: what moved, why, and what you will do next. The client wants a decision to make, not a dashboard to decode.

Section	What it covers	The metric
In one sentence	The month’s headline, in plain words	Visibility and its trend
Against competitors	Who is gaining or losing ground	Share of voice
Sources	The sites that matter, won or lost	Citation share
Tone	What answers say about the brand	Sentiment, with excerpts
Actions	What was done, what comes next	Dates, tied to the curves

Always quote excerpts from real answers: a ChatGPT sentence that names the brand, or names a competitor instead, is more convincing than a percentage. And state the period and the engines measured, so that two reports stay comparable.

Agree on the format at kickoff: who receives the report, on what date, and the three numbers you will always show, in the same order. A client who finds the same page every month spots the changes at a glance; a client who gets a new layout each time ends up reading only the last line.

The actions to offer your client

Tracking is only worth something if it leads to action. Offer actions in the order they pay off: first what blocks, then what is missing, then what takes time to build.

Unblock access: robots.txt, firewall, pages invisible without JavaScript. Our article on robots.txt lists the crawlers to let in.
Fill the missing answers: one page per important question, answering it in the first sentence, with prices where possible.
Win the cited sources: comparisons, specialist media and discussions, with verifiable information. The method is set out in our article on getting recommended by ChatGPT.
For a local client: business profile, attributes and reviews, covered in our articles on the profile and on reviews.

Date every action in your tracking: at the next report you can show what it changed, or say honestly that it has not moved anything yet.

Explaining why answers vary

Tell the client from day one that AI answers change from one day to the next and that a single check proves nothing. That is what stops a screenshot from their CEO from undoing everything. Put the explanation in writing in your first report, so you do not have to repeat it every month.

The point is easy to show: ask the same question twice in front of them. Then explain that you measure a frequency across many answers, which SparkToro considers a reasonable metric^[1]. Never promise a position: nobody controls an assistant’s answer, and an agency that claims to is exposed the first time a check says otherwise.

The affiliate program

If you recommend iPand to your clients, the affiliate program pays you a share of what they pay: 30% of the amount excluding tax, for 3 months, paid monthly^[5].

It is open to agencies, SEO or GEO consultants and content creators; the full terms are on the affiliate program page. You do not have to use it to track your clients: a regular agency account, with one subscription per brand, works the same way.

To show a prospect where they stand before any proposal, the free test asks ChatGPT, Gemini, and Perplexity a few of your customers’ questions. It is often the best way to open the conversation.

app.ipand.io/marche · fictional example

iPand

MJ Meeple Junction meeplejunction.example

SM Sarah Mitchell Meeple Junction LLC

Market

Your ranking against 6 competitors in Denver.

Market United States

Period 30 days

Type Brands

Next collection run in 4h 12m

#1 Tablewright Games

7.2% share of voice

Visibility 41% · Cited by 3 engines mentions

#2 Card Canyon

5.9% share of voice

Visibility 33% · Cited by 3 engines mentions

#3 Owlbrook Games

5.3% share of voice

Visibility 30% · Cited by 2 engines mentions

RankPlayerVisibilityShare of voiceCited byRelationship

#1 Tablewright Games 41% 7.2% 3 engines Competitor

#2 Card Canyon 33% 5.9% 3 engines Competitor

#3 Owlbrook Games 30% 5.3% 2 engines Other

#4 Mesa Peak Games 24% 4.4% 2 engines Competitor

#5 The Rolling Mug 21% 3.8% 2 engines Competitor

#6 Meeple Junction 18% 3.3% 2 engines My brand

Excerpt from the Market screen · fictional example

In the app, “Google AI Overviews” shows as “AI Overviews” when space is short.

Key takeaways

For each client, track their presence in answers, the competitors cited instead and the sources used.
Choose prompts from their customers’ real questions, branded and unbranded, and keep them stable.
A one-page monthly report: what moved, why, and what comes next, with excerpts from real answers.
Explain answer variability from day one, and never promise a position.
In iPand, one subscription tracks one brand; your clients’ subscriptions are combined in the same account.

Any tool that gives a “ranking position in AI” is full of baloney.

— Rand Fishkin, SparkToro

Frequently asked questions

How many prompts should I track per client?

Enough to cover their main intents in each of their markets. A few dozen well-chosen questions beat hundreds picked at random.

Can one iPand subscription track several clients?

No. One subscription tracks one brand. Each client has its own, and the subscriptions are combined in the same agency account, with a brand selector.

Should I track every engine?

Not necessarily. Track the ones your client’s customers use: most often ChatGPT and Google AI Overviews, then Perplexity, Gemini or Copilot depending on the audience.

How should I bill GEO tracking?

As a recurring service: the tool, analysis time and the monthly report, plus the actions you carry out. The tool’s cost depends on the number of prompts and engines tracked.