Perplexity is the only AI platform that shows you directly how many people landed on your site after reading an answer that cited you — and it's a free dashboard that almost nobody has set up yet. Ignoring it means missing the only concrete, attributable AI metric available today without paid tools. Setting it up takes half an hour and can completely change how you assess the return on your AI visibility efforts.
Perplexity is currently the only generative AI engine that gives you referral traffic readable in GA4, with `source = perplexity.ai`, and therefore attributable by landing page and by conversions. How many Italian SMEs are actually measuring it? Almost none. Here I’ll show you how to turn GA4 into an operational dashboard to understand which of your contents get picked up by AI answers and which don’t.
Let me tell you up front where I’m taking you: you don’t need a complicated setup. All you need is GA4, a well-built filter and half an hour of work to get your first useful dashboard on Perplexity traffic. Then, once you have the numbers, you’ll know where to act.
Why Perplexity is the only AI engine you can actually measure
ChatGPT, Claude and Gemini, when they answer a user, rarely generate a click toward the original site. When they do, the referral often vanishes into thin air. Perplexity, on the other hand, generates hyperlinks within its answers, and those links land straight in GA4 with `source = perplexity.ai`.
This changes everything: for the first time you have a quantitative proxy for your visibility in AI answers. Not an inference, not an estimate — real sessions, real pages, real time.
What to measure when you measure Perplexity
In the world of research on Large Language Models there’s a theme that also applies to reading Perplexity traffic: models are not neutral with respect to places, languages and entities. Manvi et al. (2024), in the paper Large Language Models are Geographically Biased, document that LLMs show systematic errors in geospatial predictions, favoring some areas and penalizing others.
From this follows a very practical operational deduction: if your business is in a territory that is less represented in training data (a minor province compared to a regional capital), it’s reasonable to expect that citations in AI answers won’t be distributed evenly. Reading Perplexity traffic in GA4 then becomes your first tool to understand whether that bias is cutting you out of citations or whether you’re appearing anyway. Without data, you’re speaking from gut feeling.
Don’t isolate the source in the filter. `perplexity.ai`, `www.perplexity.ai` and sometimes `perplexity.com` end up as three different sources.
The GA4 dashboard you can build in 30 minutes
Open Google Analytics, go to Explore → Free form exploration and create a new report. The settings you need are simple:
- Filter: `Session source` contains `perplexity`
- Dimensions: landing page, exit page, device
- Metrics: sessions, users, pages per session, average duration, conversion events
Save the report as “Perplexity Traffic”. Now you have a filtered view of only the landings that come from Perplexity’s AI answers. The pages at the top of the list are the ones the engine is citing most: they’re your current asset of citation worthiness.
If you want a second layer, pair it with Google Search Console to compare the Google queries on the same landing pages with the Perplexity sessions. You’ll often discover that the pages Perplexity cites are not the ones that rank best on Google: they’re more structured pages, with sharp definitions, which I’ve already described in the articles on the inverted pyramid and on E-E-A-T for AI.
Sort the landing pages by descending sessions and isolate the top 5.
The case study: a fish restaurant in Senigallia (AN)
Let me bring you a recent case, anonymous for obvious reasons. A fish restaurant on the seafront in Senigallia, in the province of Ancona, mid-to-upper tier, average check 70 euros. The owner calls me because he wants to understand why, over the last few months, he’s getting customers who say “I found you on Perplexity” — a new phrase, never heard before.
The intervention was simple: I set up a GA4 dashboard dedicated to Perplexity traffic, activated the conversion events “click on the phone number” and “click on the booking form”, and cross-referenced them with the landing pages.
The numbers after 60 days of observation, on a sample that holds only for that single restaurant (an indicative test, not a study):
- Sessions from Perplexity: 184 in 60 days, against 11 in the same period the year before
- Most cited pages: the “fish tasting menu” page and the blog article “how to recognize fresh Adriatic fish”
- Click-through rate on the phone number: 7.6% on Perplexity traffic, against 2.1% on generic Google traffic
The two cited pages had one element in common: sharp definitions in the first 200 characters, a bulleted ingredient list, and an author with a signed bio (the restaurant’s chef, indexed as an entity with clear author entity recognition on the Italian Wikipedia and on the official site).
The interesting jump was in attributable revenue. The 14 online bookings that the system tracked as coming from Perplexity traffic generated an estimated value of around 2,000 euros — not a step change for the restaurant, but a previously nonexistent channel that is now measurable and therefore optimizable. That’s exactly the key word: measurable.
Without the dedicated GA4 dashboard, that channel was off the radar. The bookings showed up as “direct” or “organic” because of the clicks that go from the engine to the brand to the manually typed site. With the dashboard, the pattern became visible: certain pages produce citations, certain others don’t, and you can see where to invest your next editorial efforts.
Honest caveat: 60 days are not a trend, and one restaurant doesn’t represent the category. But the pattern of cited content confirms what I see in other verticals: Perplexity prefers scannable content, with named entities and a clear authorial source.
The mistakes I see most often when people look at the dashboard
Over the last few months I’ve seen four recurring mistakes among restaurateurs and SMEs on the Marche coast who try to read Perplexity traffic without a method:
- Looking only at total sessions. They’re few by definition. The value is in the quality: pages per session, time, conversions. Almost always higher than the site average.
- Not isolating the source in the filter. `perplexity.ai`, `www.perplexity.ai` and sometimes `perplexity.com` end up as three different sources. Use `contains “perplexity”` not `equals`.
- Confusing referral and direct. Part of the AI traffic arrives as direct because the user reads the answer and then types the domain. For that you need other proxies, which we discuss in the series on backlinks as citation proxy.
- Not connecting the cited pages to the content. The dashboard tells you what Perplexity chooses, but if you don’t take those pages apart to understand why they’re chosen, you have data without action.
What changes if you have the dashboard and your competitors don’t
On the Marche coast I’ve counted, informally, about twenty mid-to-upper tier fish restaurants between Senigallia, Ancona and Numana. Only two have a GA4 setup that lets them read Perplexity traffic. The others receive it — because Perplexity cites them anyway — but they don’t see it and don’t measure it.
From this follows a very practical operational consequence: whoever measures optimizes, whoever doesn’t measure flies blind. It’s not a magic factor, it isn’t enough on its own to bring you customers, but it’s the foundation for every subsequent reasoning about the return on your visibility in AI answers.
The other proxies you need, and that you can’t read in GA4, I’ve described in the articles on event entity speaking authority and on Google Knowledge Graph entry: the Perplexity dashboard tells you how much you’re collecting, the entities tell you how much more you could collect.
What to do in the next 7 days
Three operational steps, in this order:
- Open GA4 and create the “Perplexity Traffic” report as I described above. 30 minutes.
- Sort the landing pages by descending sessions and isolate the top 5. They’re your current asset within AI answers.
- For each one, check: is there a sharp definition in the first 200 characters? Is the author a recognizable entity? Is the content structured as an inverted pyramid?
If you answer “no” to one of these points on a page that’s already being cited, optimize it: it will probably double its citations over the next 30 days.
For a more serious analysis — comparison with the 3-5 competitors Perplexity cites in your sector, multi-touch attribution between Perplexity and direct, query reverse engineering — the real analysis requires professional tools and a few weeks of data.
From the dashboard to the rest of the series
Measuring Perplexity is the first concrete step to understand whether you’re in AI answers or not. But it’s only one slice of the puzzle: ChatGPT doesn’t give you referrals, Gemini doesn’t either, and for them you need different proxies that we’ll see in the next articles in the series on how to measure AI visibility. The thread is always the same — understanding whether your brand gets picked up when the user asks the engine, and how much.
If you’re interested in digging into the bigger picture, start with how AI engines think at the tokenization level and with the weight of implicit references that stay invisible in GA4 but move citations.
