Showing clients AI-referred traffic next to visibility

Visibility charts impress clients for about two reports. Then comes the question every agency hears: "That's nice. But is it sending anyone to the site?" If the visibility numbers and the traffic numbers live in different tools, you answer by hand-building a comparison, and the comparison is different every month.

We now put the two side by side on the same page. This post explains how, what the pairing shows, and where it quietly misleads if you are not careful.

Why the pairing matters

Visibility measures whether AI answers name the client. Traffic measures whether anyone clicks through. They are related, but they can move in opposite directions. A brand can be named in a growing share of answers while few answers include a link. Or a single cited page can drive most of the visits while the wider visibility score barely changes.

Reading only one gives you a wrong story. Visibility without traffic can look like progress that does not turn into sessions. Traffic without visibility can look like a lucky page. Together they tell you whether the client is being named, where the clicks come from, and which pages earn them.

The manual way

You can do this with a spreadsheet and your analytics tool. Pull monthly AI-referred sessions by engine, using the referrer or UTM source in your analytics. Pull the visibility figure and the share-of-voice figure from your tracking. Put both in one table with the same month on each row, and add a column for what the team shipped that month.

Two cautions from our own reports.

First, referrer-attributed visits are a measured floor. Some AI answers pass no referrer, and a person might click later through a branded search or a direct visit. We say so on the page. Part of that is a platform behaviour that changed this autumn. According to the Promptwatch changelog, ChatGPT stopped sending a referrer on October 8, and visits looked like direct traffic until Visitor Analytics gained a toggle to include UTM-attributed ChatGPT visits on October 16. If a report covers that period, we note it so nobody reads a dip as a fall in demand.

Second, conversions belong to the client's analytics and CRM. We show AI-referred sessions and, where the client's system can segment AI referrers, conversions from that segment. We do not guess at revenue.

What goes on the page

Our paired page has four blocks and no more.

1. Visibility by engine. Visibility and share of voice for the client against the named competitor set, split by the engines monitored.
2. AI-referred visits by engine. Sessions from ChatGPT, Perplexity, Gemini, Claude, Copilot and the rest, with the trend line.
3. Top landing pages for AI visits. This is the bridge between the two blocks above. It shows which pages receive the visits.
4. What we shipped. A short dated list of changes in the period.

We keep commentary to a few lines under each block. A good test: if the client reads only the headings and one sentence under each, do they know what happened?

What the pairing can reveal

Some patterns we look for.

• Visibility up, visits flat. The brand is named more often, but not linked. We check whether citations to the client's own pages are changing and whether the pages cited are worth clicking.
• Visits up on one page, visibility flat. A single page is carrying the load. That is a content-concentration risk and a good place to build adjacent pages.
• Visits down, visibility steady. First check whether the referrer behaviour changed. Then check crawler errors on the landing pages, since a page that stops being fetched can stop being cited.
• Both up. Still look at which pages and prompts drove it before claiming credit.

None of these patterns proves cause. They tell us where to investigate.

Where Promptwatch fits

Promptwatch is the platform we run client programs on, and the main reason is that it keeps these two measurements in one place. Its visitor analytics tracks AI-referred traffic and conversions across ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek, Kimi, Mistral, and more, through a lightweight script, a Google Tag Manager template, or the WordPress plugin. The visibility, citation, and crawler data sits in the same project.

For agencies the plan question has a concrete answer. The changelog states that agency plans now include visitor analytics, with 500K monthly page hits on Kick-off, 2M on Growth, and 15M on Scale, so you can see how much traffic AI search sends to each client's site next to their AI visibility. That means the paired page does not depend on a separate analytics purchase for each client. Those limits are page hits on the tracking script, not sessions, so for a high-traffic client we check the hit volume against the allowance before enabling it.

On brand plans, the visitor event allowances in the fact sheet are 200K on Essential, 1M on Professional, and 10M on Business. Look at promptwatch.com for the current figures before you scope a client.

We add the same caution here as elsewhere: the platform measures identified AI referral visits and the conversions it can see through its script. It does not replace the client's CRM, and we do not present its conversion numbers as booked revenue.

Making the page repeatable

The real gain is repeatability. If this page is built from the same dashboard widgets every month, the client learns to read it. We hold the layout steady for at least a quarter. When we must change a metric definition or the competitor set, we put a one-line note on the page and keep the old definition visible for a period.

A last practical point: send the page with a short written summary, not only a link. Clients open dashboards when they have a question. They read an email when they have a minute. Two sentences that say what the page shows get the report read.

What not to claim

We do not tell clients that visibility caused the traffic, or that more visibility will produce more traffic at a fixed rate. We do not compare AI-referred conversion rates to other channels unless the client's own data supports it, and we do not estimate the share of AI influence that never carried a referrer.

Used with those limits, the paired view gets the question "is it sending anyone?" answered in the same meeting where you present the charts.