How Agencies Report AI-Referred Traffic and Conversions to Clients in 2026
A how-to for agencies: define AI-referred traffic, catch ChatGPT, Perplexity, Gemini, Claude and Copilot visits, split out paid ChatGPT clicks, and present it monthly.
Visibility scores get a client interested. Traffic and conversions keep them paying. At some point every AI search retainer hits the same question from the client's finance side: how many people did ChatGPT actually send us, and did any of them buy? An agency that can't answer that with a clean monthly number ends up defending the retainer with share-of-voice charts, which is a weak position.
This guide walks through how to set that number up per client and how to present it. It is written for agencies running several clients, so each step is something you repeat per account.
What counts as AI-referred traffic
Three different things get lumped together, and the report only works if you keep them apart.
AI crawler fetches are bots. GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and the rest request pages to train, index, or answer a question in real time. They are not visitors and never convert. They belong in a separate crawl report.
Organic AI referrals are humans who clicked a link inside an AI answer and landed on the client's site. This is the number the client is asking about.
Paid AI clicks are humans who clicked a sponsored placement. ChatGPT shows ads on its Free and Go plans in 2026, per OpenAI's help center. If a client runs ChatGPT ads, those clicks will look like ChatGPT traffic unless you separate them.
How each assistant shows up
ChatGPT is the easy one. OpenAI documents that ChatGPT adds utm_source=chatgpt.com to links it sends traffic through, so any analytics setup that keeps query parameters can segment it. The common failure is on the client's side: a redirect that drops the query string, after which those visits look like direct traffic. Test one link from a ChatGPT answer to each client's site before you trust the segment.
Perplexity, Gemini, Claude and Copilot mostly identify themselves through the referrer, when a referrer is passed at all. Expect hostnames like perplexity.ai, gemini.google.com, claude.ai or claude.com, and copilot.microsoft.com, but build the segment from what actually appears in each client's referrer report rather than a fixed list, because hostnames change. Clicks from native apps or in-app browsers can arrive with no referrer, and then they are counted as direct. No tool recovers every one of those.
Why GA4 alone falls short
GA4 can see most of these visits. The problem is where it puts them. Check how each client's property groups these sources. In many setups AI assistants sit inside the generic Referral channel next to partner sites and directories, or split between Referral and Direct, so nobody looks at them as one line. You can build a custom channel group per property, but across ten clients that is ten regex lists to maintain.
The second gap is context. GA4 knows a session came from perplexity.ai. It doesn't know which prompt produced the answer, which page was cited, or whether OpenAI's crawler fetched that page the day before. That missing join is the part the client actually wants explained.
Step 1: install tracking per client
In Promptwatch, each client is a project, and agency plans include unlimited projects. Visitor analytics goes in through a lightweight script, a Google Tag Manager template, or the official WordPress plugin. The GTM route is usually fastest for agencies, because you are probably already in the client's container.
Once it is live, visitor analytics attributes visits to ChatGPT, Perplexity, Gemini, Claude (claude.ai and claude.com), Microsoft Copilot, DeepSeek, Kimi, Mistral and other AI sources, with top pages, top locations, device and browser data, and conversions. Use the ignore-known-IPs filter to drop your own team and the client's office, or your QA clicks will inflate a small number. Visitor event limits are set per plan: 500K on Kick-off, 2M on Growth, and 15M on Scale. A high-traffic store client can use a large share of that, so check it before onboarding.
Step 2: tie visits to the pages and prompts that cited them
This is where the report stops being a traffic chart. For each landing page that received AI visits, you want the prompts that cite it.
Citation analytics lists the pages and domains each engine cites for the client's tracked prompts. Page Tracker monitors specific URLs, the client's or a competitor's, and shows the prompts and responses that cite them. Put the top AI landing pages into Page Tracker and you can say "this guide was cited in these answers and received these visits."
360 Insights puts crawler logs and citation clicks in one view, and the crawl-to-citation path shows which crawled pages actually get cited. Search Insights overlays Google Search Console and Bing Webmaster traffic per page, so you can show a page that is losing Google clicks while gaining AI citations, or the reverse.
Step 3: separate paid ChatGPT clicks
If the client advertises in ChatGPT, tag every ad destination URL with your own utm_medium and a dedicated utm_campaign, and filter those sessions out of the organic segment. Keep the campaign's spend and results in OpenAI's Ads Manager, which reports on an advertiser's own campaigns.
What Ads Manager won't show is everyone else. Ads Radar, included on every Promptwatch agency plan, captures sponsored ads in ChatGPT Search, Google AI Overviews and Google AI Mode for the client's tracked prompts, with advertiser name and domain, ad copy, target URL, and the triggering prompt. It doesn't cover Perplexity and doesn't report the client's own campaign results. It answers "who is buying ads against our prompts," which pairs well with the paid line in the report.
Step 4: present it monthly
A structure that works for most clients:
- Organic AI-referred visits and conversions by source, against last month.
- Top AI landing pages, each with the prompts that cite it.
- Paid ChatGPT clicks on their own row, from the ad campaign UTMs.
- Crawler activity on the pages that matter, including errors.
- Two or three actions for next month.
For delivery, the Looker Studio connector is free and comes with templates for monitors, citations, prompts and visibility, so you can drop AI data into the dashboard the client already opens. If the client prefers a document, scheduled PDF reports deliver default or custom dashboards on a schedule, and white-label PDFs carry your agency's branding. CSV and Excel exports cover the raw data.
Keep the language honest. Say "visits we can attribute to AI assistants," not "all AI traffic," because some of it arrives as direct and stays there.
Other tools that document AI referral traffic
If a client already runs one of these, it may cover part of Step 1. Each claim below is from that tool's entry in our directory.
| Tool | What its listing documents |
|---|---|
| AthenaHQ | GA4, Shopify, and Webflow revenue attribution |
| Wildcard | AI referral sessions, add-to-carts, checkouts, and revenue by platform, first and last click |
| Amplitude AI Visibility | AI-referred traffic tied to funnels, conversions, and revenue in Amplitude Analytics |
| AIrketing | GA4 connection that attributes sessions, events, and conversions to prompts |
| Surva AI | AI referral tracking through a JavaScript snippet; conversion tracking not documented |
| Known Agents | Human visits from ChatGPT, Claude, Gemini, Copilot, and Perplexity; conversions not documented in detail |
| Cloudflare AI Crawl Control | AI referral sources on paid Cloudflare plans; no conversion tracking |
| Attrifast | AI traffic detection with Stripe and Shopify revenue attribution |
Most of these stop at the visit or the order. Few of them also hold the prompt, the citation, and the crawl for the same page.
The recommendation
The deciding factor in AI traffic reporting is the join, not the visit count. GA4 can count perplexity.ai sessions. What a client pays an agency for is the next sentence: which answer sent them, which page it cited, and whether that page is being crawled cleanly. On Promptwatch, visitor analytics installed through the GTM template sits in the same project as citation analytics, Page Tracker, and crawler logs, and the Looker Studio connector or a scheduled PDF delivers it without a manual export. Set up one client on a Kick-off trial at Promptwatch and build the five-row report above from it, or read the full Promptwatch review.