How do agencies track client brands across ChatGPT, Gemini, Claude and Perplexity?

Quick Answer: Agencies track client brands by giving each its own prompt set and metrics, running them on a schedule, and reporting visibility and share of voice every period. If two clients would share one prompt list, you are measuring the category, not the client. This is for agencies and consultants who report AI answer visibility to more than one client brand.

Key takeaways

• Add a cadence column to every client table so weekly engines are never read day by day.
• Lock each client's prompts for the quarter, and test new wording in a group of its own.
• Before you compare tools, count the brands and prompts each client needs. Treat seats as a separate line item.
• Keep customer, ranking and traffic outcomes out of the proposal, and pitch the report itself.

How do agencies track client brands across ChatGPT, Gemini, Claude and Perplexity?

Agencies track client brands across these assistants by holding three things steady for each client: the prompts, the engines and the metrics. When those stay the same from one period to the next, a change in the report means the answers changed.

Checking by hand stops being repeatable across a roster, so agencies run the work through a tracking tool.

Welt Analytics is one such tool. It checks ChatGPT, Gemini, Google AI Overviews and Google AI Mode daily, checks Claude and Perplexity weekly, and reports a brand's visibility, share of voice, position, sentiment, sources and competitors in AI answers.

Settle one thing before you budget: tools count prompts differently. Semrush's prompt-tracking guide states: "In Semrush, the allowance is shared across all of your campaigns, and each campaign covers one platform" [1]. Ask every vendor whether a prompt counts once, or once per engine.

What should an agency track for each client brand?

Track six things for each client brand, and define each one in the same words in every report:

• Visibility: the share of the client's prompts where the brand appears in the answer at all.
• Share of voice: how often the brand is named, compared with the competitors named for the same prompts.
• Position: where the brand sits in the answer when it does appear, for example first in a list or further down.
• Sentiment: whether the answer describes the brand favourably, neutrally or negatively.
• Sources: the pages the assistant cites when it answers.
• Competitors: the other brands named beside the client.

Competitor share of voice is usually the first number a client asks about. This guide explains how to measure share of voice for a brand in AI answers versus competitors.

Daily and weekly engines in one report

Welt Analytics checks ChatGPT, Gemini, Google AI Overviews and Google AI Mode daily, and Claude and Perplexity weekly, so the client report has to show each engine's cadence.

Read a weekly engine against its own previous week. A daily engine that dips for a day and recovers tells you less than a weekly engine that drops two weeks running. Semrush notes that "The same prompt can produce different responses over time, so a brand mention that keeps showing up is stronger evidence than a single result" [1], so judge a client's presence by repeated mentions, not one run. For the monthly client report, roll the daily engines up to weekly figures so all six rows share a time unit.

How do you set up several client brands without mixing their data?

Each client gets its own prompt set, its own competitor list and its own report, and no two clients share any of the three.

The common shortcut is one generic prompt list reused for every client in the same vertical. It measures the market instead of the client. Semrush's prompt-tracking guide warns that "A category-heavy prompt list can also make a brand look stronger than it is." [1] The fix is to write each client's prompts from that client's own buyer questions and to name that client's own competitors.

The second trap shows up a month later. A new product line or a client request makes you want to edit the prompt set, and then you compare the quarter before and after. The numbers will move, and you cannot tell whether the answers changed or the questions did. Do not edit a live prompt set mid-quarter. Freeze it for the period and add new prompts as a separate group. Semrush describes the same practice: exploratory prompts sit apart, so the baseline does not shift [1].

If you are onboarding someone this week, the steps to set up a GEO tool for a new client in one day cover the account side. Before the first report goes out, run this check:

1. Load the client's frozen prompt set and its competitor list into the tool.
2. Run the set once on each engine you plan to report.
3. Open the answer for every prompt and note which brands it names.

Expected result: each prompt returns an answer showing which brands are named, so the first report has data behind it.

If an answer names competitors but never the client, keep the prompt: that is a real finding.

Plan limits to check before you take on a client

Three limits decide whether a tool fits an agency roster: how many brands it tracks, how many prompts it runs, and how many people can log in. Welt Analytics lists its Starter plan at €89 a month with 50 prompts and 1 brand, and its Pro plan at €499 a month with 400 prompts and 3 brands [2].

Say you take on three clients with 40 prompts each. That is 3 brands and 120 prompts. Starter holds one of those brands; Pro's 3 brands and 400 prompts cover the example if a prompt counts once across engines, so confirm that first.

The easy mistake is to treat seats and brands as one allowance. Vendors often sell them separately. Semrush's knowledge base says that adding a Brand Performance domain "does not grant another user access" [3]. The reverse also holds: a plan with room for five teammates may still track only one client. Count brands and prompts per client first, then ask about seats. Coupler's roundup of agency tools describes the Ahrefs Enterprise tier as offering "unlimited access to features, users, and credits" [4]. Seat terms differ by vendor and plan, so get them in writing for every tool on your shortlist.

Ask each vendor:

• How many client brands does the plan track, and what does one more cost?
• How many prompts, and does a prompt count once or once per engine?
• How many users, and can a client have view-only access?

What do you tell a client about getting mentioned in AI answers?

Tell the client what the data shows: where the brand appears, which sources the assistants cite, and which competitors are named in its place.

Clients will ask whether a ChatGPT mention brings new customers. A visibility report shows mentions, positions and sources. It does not show who bought. Report the visibility, and leave promises about customers, rankings and traffic out of the proposal. When the conversation turns to rivals, you can show a client which competitors AI assistants recommend over them with the same data.

If the client also wants content, Welt Analytics publishes new articles on the customer's blog, and every article belongs to the customer, who can export it at any time [2].

To see a client's visibility, share of voice, position, sentiment, sources and competitors across these engines, look at Welt Analytics.