How to Create an AI Traffic Report (Free Template)

Your traffic from ChatGPT and Perplexity is growing, and you almost certainly cannot prove it. Most analytics setups file those visits under Direct, or scatter them across a dozen unlabelled referrers.

That leaves you guessing in the one conversation where guessing hurts — the monthly review, when someone asks whether the AI work is paying back.

This guide walks through building an AI traffic report that holds up: which referrers to count, how to segment them in GA4, the five metrics worth reporting, and where the numbers stop being trustworthy.

Why AI Traffic Is Hard to See in Analytics

AI assistants break the assumption every analytics tool is built on: that a visit arrives carrying a referrer that names where it came from.

Some do send one. A citation clicked inside ChatGPT's web interface usually arrives with chatgpt.com attached. But plenty of paths lose it entirely.

• Desktop and mobile apps often open links in an in-app browser that passes no referrer, so the visit lands as Direct
• Copy-paste behaviour is common — people read an answer, then type or paste your URL separately, which severs the connection completely
• Referrer policies on the AI platform's side can strip the path or the whole header
• Enterprise browsers and privacy extensions remove referrers as a matter of policy

The practical consequence: your AI traffic report will always undercount. Every number in it is a floor, not a total.

That is worth stating on the first page of the report, in writing, before anyone treats a small number as a verdict on the whole channel.

Step 1: Identify Which Referrers Count as AI

Before you can segment anything, you need the list of domains. This is the part most teams get half-right — they cover the obvious three and miss the rest.

Two traps here.

Google AI Overviews do not appear as a referrer. A click from an AI Overview arrives as ordinary Google organic search, indistinguishable from a blue-link click. If someone asks you to report on AI Overview traffic from GA4 alone, the honest answer is that it cannot be isolated there — checking whether your pages appear in those answers needs an AI Overview tracker that queries the surface directly, not an analytics tool that counts visits.

Bing Copilot and Bing search blur together. Depending on the surface, a Copilot citation may report as bing.com. Decide which bucket it belongs in and document the choice, so the number means the same thing every month.

Step 2: Build the Segment in GA4

GA4 will not group these for you. You have two routes, and the difference matters more than it looks.

1. Create a custom channel group. Admin → Data display → Channel groups → Create new. Add a channel called "AI Assistants" and place it above Organic Search and Referral in the ordering, since channel rules resolve top-down and a lower rule will swallow your traffic.
2. Define the match condition on Source, using a regex that covers every domain from Step 1:

^(chatgpt\.com|chat\.openai\.com|perplexity\.ai|.*\.perplexity\.ai|gemini\.google\.com|bard\.google\.com|copilot\.microsoft\.com|claude\.ai|you\.com|poe\.com|phind\.com)$

3. Build a comparison instead, if you want history. Custom channel groups apply going forward and to the last 30 days only. For longer look-backs, build the same regex as a comparison in a report, or as a segment inside an Exploration.
4. Verify against raw data. Reports → Acquisition → Traffic acquisition, set the dimension to Session source, and search for perplexity. If the row exists but your channel shows zero, the ordering is wrong.

The channel-group route is better for recurring reporting because the grouping persists and anyone can use it. The exploration route is better for the first month, when you need to look backwards to establish where you started.

Set a baseline before you change anything. A month-one number with nothing to compare it to is a statistic, not a finding. Pull the previous 90 days as your reference point on day one.

Step 3: Choose the Five Metrics That Matter

Session count is the number everyone asks for and the least useful one in the report. It moves with the AI platforms' own growth, which has nothing to do with your work.

These five earn their place:

• Sessions by engine — not a single total, but the split. A jump concentrated in one engine usually means you got cited on one popular prompt, which is a finding you can act on.
• Engagement rate against your site baseline — AI referrals tend to arrive further along in their thinking, so engagement above your site average is the signal that this traffic is qualified rather than accidental.
• Landing page distribution — which pages get cited. This is the most directly actionable number in the report, because it tells you what kind of content the engines pick up.
• Conversion rate against organic — the comparison that answers "is this traffic worth anything", and the one that makes the case for further investment.
• New vs returning split — heavily new-weighted AI traffic means genuine discovery rather than existing customers taking a different route in.

Report the engine split as a share, not just counts. Share of your AI traffic by engine is stable enough to trend; raw counts move with platform growth and will tell you a story about OpenAI's user numbers rather than about your content.

Step 4: Structure the Monthly Report

A report that opens with a chart makes people work out what it means. A report that opens with a sentence tells them, and uses the charts as evidence.

Four pages, in this order:

• Page 1 — The answer. Three numbers: AI sessions this month, change against last month, and conversion rate versus organic. One sentence under them saying what happened and why.
• Page 2 — Where it came from. Engine split as a share, trended over the months you have. This is where a single-engine spike becomes visible.
• Page 3 — What got cited. Landing pages ranked by AI sessions, with engagement rate beside each. This page is the content brief for next month.
• Page 4 — Caveats. The undercount explanation from the top of this guide, plus anything unusual in the period — a tracking change, a big content launch, a platform outage.

Page 4 is the one people skip and the one that protects you. When a number moves oddly three months from now, a documented caveat is the difference between a quick explanation and an afternoon of reconstruction.

Step 5: Build It in Looker Studio

Once the GA4 segment exists, the dashboard is mostly assembly.

1. Connect GA4 as a data source — the native connector, not a BigQuery export, unless you already have that pipeline running.
2. Add a report-level filter on Session source using the same regex from Step 2. Putting it at report level rather than per-chart means one edit updates every page when a new engine appears.
3. Add a date-range control and a comparison set to previous period, so every chart shows movement without anyone building a second view.
4. Break out engine as a dimension on the page-2 charts, using Session source grouped into a calculated field, so the seven domains collapse into five readable engine names.
5. Set the refresh cadence to daily. Hourly wastes quota on data that moves in weeks.

One trap worth naming: do not blend AI traffic with organic in a single chart with a shared axis. The volumes differ by an order of magnitude, which flattens the AI line into the baseline and makes a genuine trend invisible. Separate charts, or an indexed view.

Budget two to three hours for the first build, and expect to revisit it. The referrer list is the moving part — a new assistant launches, a domain changes, and your regex quietly stops matching. Diarise a quarterly check, or the report degrades without ever looking broken.

The Template, Two Ways

Everything above is the template. The referrer list, the regex, the five metrics and the four-page order are the whole specification — copy them, build it once, and it is yours. Nothing in this guide is gated.

The second way is to not build it. Easy Fetcher ships this exact report as a prebuilt dashboard, connected to GA4, with the assistant split, the share-of-sessions breakdown and the session trend already laid out. You create an account and point it at your property; there is no regex to write and nothing to maintain when a new assistant appears.

Two details in it are worth stealing even if you build your own. It states the undercount on the dashboard itself — "Analytics folds these into Referral or Direct, so they are matched on referrer hostname rather than read off a channel" — so the caveat travels with the numbers instead of living on a page nobody opens. And it can be scheduled, which is what turns a dashboard someone has to remember to check into a report that arrives.

What This Report Cannot Tell You

Being straight about the limits is what makes the rest of the report credible.

• Assisted conversions are mostly invisible. Someone who reads about you in ChatGPT on Monday and searches your brand on Thursday converts as branded organic. The AI influence is real and your report will attribute it elsewhere.
• AI Overview clicks are not separable from ordinary Google organic, as covered in Step 1.
• Month-to-month volatility is high. Prompts get answered differently as models update. Treat a single month's movement as noise until a trend holds across three.
• You cannot see the answers you appeared in — only the clicks that followed. This is the largest gap, and it deserves more than a bullet.

The honest framing for a client conversation: this report measures the traffic AI sends you, which is a subset of the influence AI has. Use it as a floor and a trend line, not as the whole picture.

The Gap This Report Leaves

Analytics can only count visits. It starts measuring at the click, which means everything before the click is invisible to it.

That missing half is where most of the outcome lives:

• Whether you were named at all for the questions your buyers ask
• Whether you were linked, or merely mentioned without a citation
• Where you sat in the answer — named first reads very differently from named seventh
• Who got named instead, and which sources the assistant leaned on to build its answer
• How you were described — a neutral mention and an enthusiastic recommendation both count as one mention

None of that produces a session, so none of it reaches GA4. A brand can be named in half the answers in its category and still show almost nothing in this report, because most AI answers resolve the question without sending anyone anywhere.

Closing that gap needs the opposite method. Instead of counting visits that arrived, prompt tracking runs a fixed set of questions on a schedule, across each assistant, and records what the answer actually said. Same discipline as rank tracking, different unit — the prompt rather than the keyword.

Run both and the picture completes. Prompt checks tell you whether you are in the answer. This traffic report tells you what happens when you are.

Where Easy Fetcher Fits

Building the GA4 side by hand is realistic — that is what this guide is for. Running scheduled prompt checks across six assistants, and recording position, citations and sentiment on each, is not something you assemble in an afternoon.

Easy Fetcher does both halves. It runs your prompt set across ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI Mode, and reports visibility, citations, position and sentiment out of 100 — against the same competitor set, so you can see who gets named in your place and which sources the assistants lean on when they do.

On the traffic side, AI traffic analytics connects GA4 and Search Console directly, and reports assistant sessions, share of total sessions, which assistant sent them and which landing page they hit. The report this guide describes, without the regex or the quarterly maintenance.

Worth saying plainly: the GA4 method above is free and yours to keep. If you are measuring one site and have the afternoon, build it. The case for a tool is the prompt-check half, which has no manual equivalent — a free demo will show you what that looks like on your own prompts, and the plans are there if it earns its place.

Conclusion

Three things determine whether this report is useful. The referrer list has to be complete, or you are measuring a fraction of a fraction. The engine split has to be reported as share rather than raw counts, or you are tracking OpenAI's growth instead of your own. And the caveats page has to exist, or the first odd number destroys trust in all the others.

Remember what it is, too. This report counts the visits AI sends you, which is the smaller half of AI's influence on your pipeline. The larger half is the answers that named you and sent nobody.

Build the GA4 channel group first and pull your previous 90 days as a baseline today — the report is only as good as the comparison you can make against it next month.