AI visibility reporting for agencies: how to offer it to your clients

The first conversation: start with the prospect's own result

Before the meeting, run the free scan with the prospect's brand name and category; adding their website is optional. The scan asks ChatGPT five questions that do not contain the business name, in the target country you pick, and shows the result in a scorecard. You see in how many answers the business was named, who was named instead, and, when sources come back, which sites the answers relied on. No signup is needed.

This changes the meeting. Instead of "AI matters", you say "in these five questions, ChatGPT named these companies instead of you". The topic stops being a trend and becomes the prospect's own market.

A few rules keep this honest:

• Show the result as it is. If the prospect was named, say so. Who else was named, and the sources, are still useful.
• Do not present one result as a verdict. The scorecard is one engine (ChatGPT) at one moment: a snapshot, not where the business stands for good. It can be created once per business every 30 days, so you cannot rerun it until you get the result you want.
• Get the prospect's permission before you show their result to anyone else or use it in ads or case studies. Do not assume the free scorecard stays with you either: there is one free scorecard per business for 30 days, and anyone who scans that business in that window is sent to the same scorecard. Branded scorecards on the Agency plan are not shared this way; each one is created in your agency account.
• The Discovery Visibility on the free scorecard is not the SOV Score. It shows how many answers named the business, not its share against competitors you chose. The SOV Score needs a competitor list that you set with the client first.

Choose the questions and competitors with the client

A measurement is only as good as its questions. Do not invent them at your desk; get them from the client, and from their sales team if they have one. Ask three things. What do your buyers ask before they buy? Who do you lose deals to? Which country and which language do you sell in?

The questions should not contain the brand name. For a furniture brand, a buyer might write:

• "Which brands make durable, affordable sofas for a small living room?"
• "What should I check before buying a solid wood dining table?"
• "Which furniture brands are reliable to buy from online?"

Then settle the rest of the scope:

• Competitors. Start with the companies the client loses deals to, then add the names the scorecard showed instead. A marketplace or comparison site in the answer may not be a rival; it may be where the buyer is sent. We saw this with real estate in our field note on local businesses.
• Country and language. A question is tracked separately for each country and language, and each of those counts as one tracked prompt. Tracking one question in Turkish and in English for Türkiye counts as two tracked prompts.
• Keep the question set stable. If you change the questions every month, the comparison breaks. The SOV Tracker report does not show a change when the previous measurement used a different question or engine scope. It does not compare what cannot be compared.

The monthly report: four things are enough

Keep the report short. Once a month, answer four questions clearly:

• Where is the SOV Score going? The SOV Score tells you this: of all the times your client and the competitors you chose together were named in the answers, what share went to your client. It comes from the weekly measurements. It carries the same name in the dashboard, the weekly email and the PDF report. Look at the direction over several weeks, not at one week.
• Who was named instead of your client? Which competitor came up on which question, and did a new name appear?
• Which sources did the answers rely on? The client's own site, or third-party sites? This gives you a clue about where to work next month.
• What did you change this month? Which page you wrote, which profile you completed, which publication covered the client. The tool does not write this part; you add it as your own note, with dates.

Illustrative example, not a measurement: the SOV Score went from 22% to 28% over four weeks. That means the client's share of all mentions among them and the competitors you chose together grew. It is not a grade; it is a share. If the competitor list changes, the share changes too.

Two rules keep the report trustworthy. First, not measured is not zero. If an engine could not be measured that week because of a technical problem or an outage on the AI company's side, the report does not count those answers as zero. It marks the engine as incomplete and labels the score as partial. If too few engines were measured, it shows "not measured" instead of a score. Do not fill that gap with a zero in your slides either. Second, do not tie a change to a single action. If the SOV Score rose after you published a page, that is a clue, not proof.

If the client asks where the numbers come from, show them our methodology. It is written for exactly that question.

How to talk about results: do not promise rankings

Nobody controls what AI assistants say. Not you, not us, not any other agency. There is no fixed "ChatGPT ranking" either; the answer changes from question to question and from day to day. So the honest thing to sell is the work and the measurement, not a promise about the result.

Often it is enough to change one sentence in your pitch:

• Instead of "We will get you into ChatGPT's recommendations", say "We will measure your questions every week on 5 engines and show you once a month what changed."
• Instead of "Your score will double in three months", say "These are the pieces of work for the next three months, and after each one we will report what happened in the measurement."
• When the SOV Score drops, do not hide it: "Your SOV Score fell this month. A new comparison article shows up in the sources; here is our next step."

Write the contract the same way. Tie your fee to the work and the regular report, not to a score target. Tying your fee to a score ties your income to something you do not control.

Where SOV Tracker Agency fits

You can do all of this with a spreadsheet and screenshots taken by hand. After a few clients it gets hard. The Agency plan takes over the repetitive part:

• A separate workspace for each client, up to 10 projects in total including your agency's own brand, all managed from one agency panel.
• You invite the client to their own workspace as a viewer. They see their data and cannot change it. Your own team joins the agency workspace; a client workspace only takes viewers. Agency owners and admins can switch into client workspaces.
• PDF reports with your agency's logo and color. The report covers the SOV Score, competitors, sources and the change against the previous measurement. A small "Powered by SOV Tracker" line stays at the bottom; there is no option to remove it.
• Branded scorecards for prospects: 20 branded scorecards a month included, then 3 credits each (credits are extra usage units you buy in the dashboard). The scorecard opens with your agency's logo, and its questions do not contain the prospect's name.
• 300 tracked prompts shared by your agency and all clients, measured every week on 5 engines. Split evenly across 10 projects, that is 30 tracked prompts each.

The report comes from the measurement, so nobody on your team collects screenshots or copies numbers into a spreadsheet. What is left for you is the fourth part: adding what you did this month, with dates, as your own note alongside the PDF.

The Agency plan is $449 a month. The full comparison is on the pricing page.