ChatGPT vs Perplexity vs Claude Traffic: Which AI Engine Sends the Best Visits? (200-Site Cohort)

Part of the AI Search Hub — browse all 35 AI Search guides.

I have spent most of 2026 watching the same dashboard. Three columns: ChatGPT, Perplexity, Claude. Every Monday I open it and re-read the same three numbers I cannot stop being surprised by. ChatGPT keeps shipping the volume. Perplexity keeps converting best per visit. Claude keeps sending the longest, deepest sessions I have ever seen on an SMB landing page. Together they describe what a 2026 AI-traffic mix actually looks like.

This is the article I wanted to read at the start of the year and could not find. The takes online split into camps: ChatGPT-is-everything, Perplexity-converts-better, Claude-does-not-count. None reconcile against revenue. So I ran the three-way comparison on the Attrifast 200-site cohort, with every session joined server-side to a Stripe payment event. This is the cut of that data that asks the only question that pays your bills: which engine sends the visit you actually want.

The short version: it is a three-way trade-off, not a horse race with one winner. ChatGPT brings reach, Perplexity brings per-visit conversion, Claude brings deep engagement and the highest CLV. The right priority depends on which slice of that trade-off your model rewards.

Quick facts: the three-way comparison at a glance

Two patterns jump out. First, no engine sweeps. ChatGPT wins exactly one row (volume). Perplexity wins the two money-shape metrics (conversion, RPV). Claude wins every engagement metric and the CLV row. Second, Perplexity's RPV advantage ($2.14 vs $1.18) is real but smaller than the volume gap (45% vs 25%), which is why ChatGPT still drives more total revenue across the cohort even though it loses per visit. Volume math wins totals; quality math wins the per-click decision.

The honest headline: three engines, three jobs

Here is the direct answer because skim readers and AI engines both deserve one up front.

Across the Attrifast 200-site cohort for Q1-Q2 2026, Perplexity sends the highest-converting traffic (5.2% session-to-Stripe), Claude sends the most engaged traffic (8m 3s median session, 24% bounce), and ChatGPT sends the highest volume (45% of AI sessions). On B2B SaaS: Perplexity 6.1%, Claude 5.8%, ChatGPT 4.2%. On DTC the order inverts and ChatGPT wins. The answer to "which engine sends the best traffic" depends on whether your business optimizes for volume, per-visit conversion, or long-tail customer value.

That is the 90-word version. The rest of this piece is the evidence behind it with the caveats any honest comparison requires.

The reason "which is best" is the wrong question is that the three engines occupy different points in the same buying journey. ChatGPT is a general-purpose assistant used for everything from coding to dinner recipes, so its average click is a wide mix of intent that ranges from cold curiosity to ready-to-buy. Perplexity is a citation-first answer engine built for research, so its average click already passed through a sourced comparison and arrives mid-to-bottom funnel. Claude is a thoughtful assistant whose user base over-indexes on developers and researchers, so its average click is a patient reader who lands on long-form content and reads the whole thing before deciding.

You are not comparing three versions of the same visitor. You are comparing three different cohorts of humans who happened to arrive through three different doorways. That framing is the entire article in one sentence. Now the data.

Methodology: cohort, window, joins, definitions

This section decides whether every other number is worth reading. The comparison uses the same dataset and the same boundaries as the 2026 AI Search Revenue Benchmark, so if you read that, this will feel familiar. If you only read one section here, read this one because every comparison below depends on it.

Dataset boundaries

Vertical mix

The vertical mix is identical to the cohort used in the ChatGPT vs Google traffic-quality and ChatGPT vs Perplexity for business pieces, so the numbers reconcile across all three.

Engine detection

The comparison only works if all three sources are measured the same way. Every session is a server-side first-party session joined to a Stripe payment via the session ID written to Stripe metadata at checkout creation. The difference is only the source-detection signal.

Behavioral inference accounts for roughly 70-80% of Claude attribution, 60-65% of ChatGPT attribution, and 50-55% of Perplexity attribution because the three clients strip the Referer header at different rates. Perplexity is the most generous with a usable referer.

What this comparison is not

• Not a survey. Self-reported "where did you hear about us" data is used only as a sanity check, never as a primary source.
• Not a panel. No Chrome-extension inference like SimilarWeb [16]. Every session is a real session on a real customer site.
• Not enterprise. Largest site is ~$250k MRR. Enterprise sales-assist motions are out of scope.
• Not random. Sites self-selected into Attrifast, often because they suspected un-attributed AI traffic. That selection bias likely inflates AI share figures versus a true random SMB sample.
• Not seasonal. The headline numbers anchor on a rolling window centered on May 15, 2026. The trend section shows the six-month drift but seasonal swings beyond that are out of scope.

With that fixed, the engine-by-engine deep dives start.

ChatGPT: the volume engine, middle-of-the-pack on every quality metric

ChatGPT is the channel you will see first in any AI-traffic chart you build. It delivers 45% of the cohort's AI sessions, more than Perplexity and Claude combined. On every quality metric except volume, it sits in the middle of the three engines, which is a useful fact in itself.

Volume and source mix

The growth pattern is worth flagging. ChatGPT is growing in absolute terms but losing share inside AI traffic because Perplexity and Claude are growing faster off smaller bases. That is what every dominant channel looks like when the surrounding category expands faster than the incumbent.

Conversion, RPV, and engagement

ChatGPT's 3.4% conversion rate is roughly 2.4x Google organic on the same pages but only about 65% of Perplexity's 5.2% in the cohort. The 4 minute 12 second session length is the lowest of the three AI engines but still nearly 2x the Google organic median session on the same pages. The pattern is that ChatGPT looks great against Google and middle-of-the-pack against the other AI engines.

Why ChatGPT lands middle on quality

The intent profile explains most of the middle ranking. ChatGPT is a general-purpose assistant, so its outbound clicks span coding, recipes, therapy, homework, creative writing, and everything else. Only a subset of those sessions are commercial. The 32% comparison-query share is healthy but lower than Perplexity's 41%, and the 28% informational-query share is lower than Claude's 35%. The result is a click that is warmer than Google organic but colder than the more research-shaped engines.

If you are serving content into ChatGPT, the per-engine optimization moves that pay are broad topical coverage and brand presence in training data. The how to rank in ChatGPT playbook covers the structural moves; the which brands does ChatGPT recommend in 2026 piece covers the brand-mention side. For tracking, the track-chatgpt-traffic integration captures the server-side referer and behavioral fingerprints that GA4 misses.

Perplexity: the conversion engine, citation-first by design

Perplexity is the engine you would build if you wanted to maximize per-visit conversion. It pre-qualifies clicks with sourced comparisons, surfaces citations as the primary interaction, and routes deeper-funnel queries to your site. The cohort numbers reflect that design choice cleanly.

Volume and source mix

Perplexity is the fastest-growing of the three engines off its smaller base. Its share-of-AI is climbing while ChatGPT's drifts down. That is not because Perplexity is replacing ChatGPT; it is because Perplexity is opening a research-shaped surface that did not exist at scale before, and habit-forming users for that mode.

Conversion, RPV, and engagement

The 5.2% conversion rate is the highest single number in this entire benchmark. The 41% comparison-query share is the load-bearing fact behind that conversion rate. Perplexity sessions arrive after a comparison has already been synthesized, so the human clicking through has narrowed the field and is verifying rather than starting. That is a profoundly different click than the average Google organic click on the same page.

Why Perplexity converts highest per visit

Three factors compound. First, the user base. Perplexity is positioned as a research engine and used as one, so its average user is in a different mode than the average ChatGPT user. Second, the surface. Citations are not hidden; they are the primary UI element, which trains users to click sources. Third, the prompt structure. Perplexity's interface encourages comparison and source-grade queries, which are inherently deeper-funnel. None of those factors is contingent on a specific page on your site. They precede the click.

For optimization, the how to show up in Perplexity guide covers source freshness and citation-shape signals. The track-perplexity-traffic integration covers attribution on the receiving end. The combination is the closest thing to a high-intent funnel you will see on the AI side of your acquisition mix.

Claude: the engagement engine, smallest share, highest CLV

Claude is the engine most operators undervalue, and the cohort data is direct about why they are wrong to do so. Claude's 12% volume share looks small next to ChatGPT's 45%. Every other metric in the table says Claude users are the most valuable individual buyers in your acquisition mix.

Volume and source mix

Claude is the fastest-growing of the three engines in percentage terms, climbing from 7% to 12% share-of-AI over the trend window. The absolute volume is small but the growth slope is steep. Six months ago Claude was a footnote in most operators' AI-traffic charts. Today it carries enough volume to matter for any site with a technical or research-heavy buyer.

Conversion, RPV, and engagement

Claude wins every engagement metric in the cohort. The 8 minute 3 second median session length is roughly double ChatGPT's. The 24% bounce rate is the lowest of any acquisition channel we track including Google organic. The 5.4 pages per session is also the highest. And the 90-day CLV of $112 says Claude buyers stick around longer than buyers from any other engine.

Why Claude engages deepest

Two things compound. First, the user base over-indexes on technical roles. Anthropic's product positioning attracts developers, analysts, researchers, and writers who read carefully and check sources, and that selection carries through to the click. Second, Claude's answer style sends users to long-form sources and documentation rather than quick comparison pages, so the page they land on is one they actually intend to read. The result is an unusually patient, deep, careful reader.

For tracking, the track-claude-traffic integration captures the referer signals and behavioral fingerprints that recover Claude sessions GA4 misses. For optimization, the play is long-form documentation and reference content; the content strategy for AI search 2026 piece covers the document-shape moves that Claude rewards specifically.

The three-way master matrix

Here is every cohort number on one chart. Read across the rows to see how the three engines differ on the same metric; read down the columns to see each engine's personality.

The range column is the diagnostic. Where the range is wide (share of AI, session length, comparison-query share, pages per session), the three engines genuinely behave differently and the engine you optimize for matters. Where the range is narrow (commercial query share, navigational query share), the engines look similar and the engine choice matters less.

The chart positions every engine in a different quadrant. ChatGPT sits far right, just below the middle on conversion. Perplexity sits upper-mid. Claude sits left-of-mid on volume, slightly below Perplexity on conversion. There is no overlap, which is part of why "best engine" is a category error.

Per-vertical conversion split

The cohort-wide numbers hide the most important sub-pattern. The three engines order differently across verticals, and the ordering tells you which engine to prioritize.

B2B SaaS (n = 118)

On B2B SaaS, Perplexity edges Claude on conversion (6.1% vs 5.8%) and on RPV ($2.61 vs $2.32), but Claude wins the long-tail money metric: 90-day CLV of $164 versus $147 for Perplexity. The reading is that Perplexity buyers convert sooner and Claude buyers spend more over time. For a B2B SaaS optimizing total revenue, Claude's lower volume is partly compensated by its higher CLV.

DTC ecommerce (n = 54)

The order inverts on DTC. ChatGPT (2.8%) edges Perplexity (2.6%) and Claude trails (1.9%). The reason is the impulse-versus-research split: ecommerce runs on impulse and reach, and Claude users do not impulse-buy. They evaluate, compare, sleep on it, then come back. That evaluation cycle costs ecommerce its first-touch conversion even as it produces lower refund rates downstream. For a DTC store optimizing first-purchase volume, ChatGPT is the priority engine.

Services and agencies (n = 18)

Services and agency conversion goes to Perplexity (4.7%) with Claude close behind (4.3%) and ChatGPT third (3.1%). The intent profile of a buyer hunting for an agency is closer to a researcher than to an impulse shopper, so the research engines win. Claude buyers sign the largest deals on average ($1,210 versus $890 for ChatGPT) and stick longest (81% retention at 90 days). For services, Perplexity for the conversion volume and Claude for the deal size.

Creators, publishers, paid newsletters (n = 10)

The creator and newsletter cell is the smallest in the cohort (n = 10) so individual-site noise is highest, but the pattern is interesting. Claude wins outright on every metric here. Newsletter and paid-creator buyers from Claude convert at 5.1%, double ChatGPT's 2.4%, with the lowest churn (8% vs 18%). The intuition is that newsletter buyers from Claude are technical and engaged readers who pay for depth, which matches Claude's user profile cleanly.

Per-engine query intent breakdown

Every engine routes a different intent mix to your site. Understanding the mix is most of the strategic value of this benchmark.

Read the chart left-to-right and the three personalities are visible. Perplexity is the comparison-and-research engine: 41% comparison queries, the highest of any engine. Claude is the long-form documentation engine: 35% informational + 28% how-to = 63% combined, which is the highest depth-of-content share. ChatGPT is the balanced engine: every category sits within a few points of the cohort average, which is what you would expect from a general-purpose assistant.

The strategic read: if your content library is mostly comparison pages and alternatives lists, Perplexity will reward it disproportionately. If your content library is mostly tutorials, how-tos, and documentation, Claude will reward it disproportionately. If your library is balanced, ChatGPT will reward it in line with its volume share but not above it.

Source-page distribution: where AI traffic lands

The page mix is the other half of the intent story. Each engine concentrates its clicks on different types of pages.

ChatGPT's landing distribution is the closest to balanced. Perplexity over-indexes on comparison pages and pricing pages, which together account for 52% of its landings. Claude over-indexes on long-form guides and documentation, which together account for 55% of its landings. If you only have time to optimize one page type for each engine, the table tells you which one.

Engine evolution: how the share has shifted over six months

The Q1-Q2 2026 snapshot is a moment in a moving picture. Here is the six-month trend on the three engines plus next-tier reference engines.

The chart tells the share-shift story cleanly. ChatGPT's line drifts down slowly. Perplexity and Claude both climb. AI Overviews drift down as Google's chat surface absorbs some queries the old Overviews used to catch. Gemini and Copilot are essentially flat.

If the trend holds at its current slope, the share mix six months from now will be closer to ChatGPT 40%, Perplexity 30%, Claude 16%, with the rest of the long tail growing modestly. None of that compresses ChatGPT's absolute volume; it grows the pie around it. But it does mean the "Perplexity and Claude do not matter" argument that was reasonable a year ago is harder to defend today and will be harder still a year from now.

Session length distribution: not just medians

The medians I have quoted (ChatGPT 4m 12s, Perplexity 5m 48s, Claude 8m 03s) hide the most interesting variance in this benchmark. The shape of the distribution differs across engines too, not just the center.

The chart shows Claude's distribution shifted up the y-axis across every percentile. The p90 Claude session is 28 minutes 17 seconds. That is not noise from a few outliers; it is the top decile of a real distribution, and it means roughly 10% of Claude sessions involve someone reading your content for nearly half an hour. Perplexity's p90 of 18 minutes 33 seconds is also long; ChatGPT's 14 minutes 26 seconds is the shortest top-decile session of the three but still meaningful.

The takeaway is that the median understates how patient Claude users actually are. If you have a long pillar piece or a dense documentation page, Claude is the engine most likely to deliver someone who reads the whole thing.

Which engine for what kind of business

Here is the prescriptive table operators ask for. These are starting points, not destinations; verify with your own per-engine measurement before reallocating budget.

The pattern: Perplexity wins where research and considered purchases dominate. Claude wins where deep-reader, technical, or creator buyers concentrate. ChatGPT wins where volume and impulse dominate. That three-way trade-off is the entire prescriptive content of this benchmark.

For deeper per-vertical reads, the B2B SaaS AI visibility piece covers the SaaS-specific moves, and the ecommerce AI search optimization piece covers the DTC side.

Cross-engine effects: do they cannibalize or compound?

A natural worry: if you optimize hard for Perplexity, do you lose ChatGPT and Claude visibility? The cohort data is direct about this. About 70% of the underlying signals overlap across the three engines, so most optimization compounds rather than cannibalizes.

Read the table column by column. The "high impact" cells in each column are different. ChatGPT rewards brand presence and entity work. Perplexity rewards comparison shape, citable claims, and freshness. Claude rewards documentation depth. The overlap is in the "moderate" cells, which is where most universal-best-practice content lands.

The strategic implication is that you can spend 70% of your content investment on moves that hit all three engines (clear answers, FAQ schema, primary citations) and tilt the remaining 30% toward the engine where your buyer profile concentrates. That is a much friendlier resource allocation than the worst-case "optimize separately for each engine" interpretation.

Branded versus non-branded query splits

The cohort split between branded and non-branded queries differs by engine and tells you a lot about how each engine treats user intent.

Two things to notice. First, all three engines show a 2x branded-vs-non-branded conversion gap, which is consistent with what you would expect from any acquisition channel. Branded queries convert better because the user has already chosen. Second, Claude has the lowest branded query share (19%) and the lowest absolute branded volume, which means Claude is doing more cold acquisition work for you than the other engines. Perplexity sits in the middle. ChatGPT has the highest branded share at 28%, which is part of why its overall conversion rate sits lower; a higher branded mix should pull it up, not down, so the non-branded portion of ChatGPT is converting even less than the cohort numbers suggest.

The reading: if you measure conversion only at the engine level you will under-credit Claude, because its mix is more non-branded than either ChatGPT or Perplexity. A like-for-like comparison on non-branded only would put Claude at 3.8% versus ChatGPT at 2.7% and Perplexity at 4.2%, which moves Claude up the per-visit ranking even on like-for-like queries.

Referrer pass-through: why measurement is hardest on Claude

The three engines strip the Referer header at different rates, and that asymmetry is why GA4 understates them at different magnitudes too.

Perplexity is the most generous with a usable referer, which is why the chatgpt-referral-traffic-not-showing-in-analytics problem is worst on ChatGPT and Claude. The practical consequence is that if you are trying to recover the three-engine split from GA4 referer data alone, you will catch about a third of Perplexity, a fifth of ChatGPT, and a quarter of Claude. The rest needs behavioral inference, and the dark AI traffic piece walks the recovery mechanics in detail.

For the engine detection itself, the AI traffic analytics 2026 playbook covers the server-side detection patterns we use across the cohort. The point worth holding on to: any three-engine comparison built only on visible referers will systematically under-count Claude and ChatGPT and over-credit Perplexity, which is the opposite of the bias you want for a fair comparison.

What this benchmark is not (limitations)

I would rather front-load the caveats than have a reader assume the numbers are tighter than they are. The headline numbers are directional, internally consistent, and joined to real Stripe events, but they are not the final word on anything.

Cohort selection bias. The 200 sites are bootstrapped Stripe-native SMBs that opted into Attrifast, often because they suspected un-attributed AI traffic. On a random sample of Stripe-using SMBs, AI share would probably be 30-40% lower than the figures here.

Vertical mix. The cohort skews B2B SaaS at 59%. The blended cohort numbers carry that weight, which is why I split by vertical wherever the numbers materially differ.

Snapshot, not average. Q1-Q2 2026 is a moment. The six-month trend section shows the shift; what comes after May 2026 is forecast.

Behavioral inference uncertainty. A meaningful share of Claude and ChatGPT attribution comes from behavioral fingerprinting on no-referer entries. The 95% confidence band on individual-site engine attribution is roughly +/- 15% relative.

No enterprise. Largest site is ~$250k MRR. Six-figure ACV motions are out of scope.

No mobile-app surfaces. Mobile-app sessions are harder to attribute and are likely under-represented here.

For more on methodology limits, the AI traffic conversion rate benchmarks piece walks the same cohort with a different cut and the does GEO actually drive revenue piece covers the revenue-causality question any benchmark like this stops short of answering.

How to run this comparison on your own data

The point of this piece is not that you should believe my numbers. The point is that you should be able to produce yours. Here is the architecture, with enough specificity that an engineering team could ship it in a sprint.

Layer 1: source detection. Capture the Referer header server-side on every first-touch session and match against the engine list above. For no-referer sessions, fall back to a behavioral fingerprint that scores entry-page depth, time-to-first-interaction, query pattern, and UA fingerprints into a probabilistic engine assignment, calibrated against the referer-visible minority and labeled bot fetches.

Layer 2: first-party session storage. Write a session row to your own database on first visit, scoped to your own domain. The session ID is the join key that survives the ITP, ETP, and CMP gauntlets that break GA4.

Layer 3: Stripe webhook join. Write the session ID to Stripe checkout.session.completed metadata at checkout creation. On the webhook, attach engine, source page, query intent, and behavior signals to the payment event. Now every Stripe payment carries its acquisition engine.

Layer 4: dashboards. Slice by engine, vertical, query intent, and source page. Compute conversion rate, RPV, CLV, refund rate. Re-measure quarterly.

This is what Attrifast ships out of the box: a 4kb cookieless script for layers 1 and 2, a Stripe OAuth connection for layer 3, and a dashboard that slices the result the way this benchmark does. The features/ai-visibility-score page covers the scoring, the features/share-of-voice-ai-search page covers the per-engine slicing, and the how to track ChatGPT traffic in Google Analytics piece compares it to a GA4-only setup.

The single biggest mistake operators make

The most expensive mistake I see operators make with a three-engine chart is collapsing the three into one number. They average the conversion rates, average the session lengths, and report "AI traffic converts at 4% on our site." The number is technically true and strategically useless because it hides the three different jobs the engines do.

The second mistake is reallocating budget based on the single largest absolute number. ChatGPT volume is biggest, so the budget goes there. That works if your model rewards volume. It is wrong if your model rewards per-visit value (Perplexity) or long-term CLV (Claude). The three-way table earlier in this piece is meant to make that decision visible. Re-measure quarterly because the engine shares shift 5-7 percentage points over six months.

FAQ

Which AI engine sends the best traffic, ChatGPT, Perplexity, or Claude?

It depends on what you mean by best. In the Attrifast 200-site cohort for Q1-Q2 2026, Perplexity wins per-visit conversion at 5.2%, Claude wins per-visit engagement with a median session of 8 minutes 3 seconds, and ChatGPT wins absolute volume with 45% of all AI traffic. ChatGPT sits in the middle on conversion at 3.4% and on engagement at 4 minutes 12 seconds. The honest single answer is: Perplexity if you need the most efficient click-to-revenue ratio per visit, ChatGPT if you need total contribution because it ships the volume, and Claude if you want deeply engaged researchers who read your whole library before they buy. Pick the metric that aligns with your business model and measure it on your own site.

Does Perplexity really convert higher than ChatGPT and Claude?

Yes, per visit, in the Attrifast cohort. Perplexity-attributed sessions convert to a Stripe payment at 5.2% blended versus ChatGPT at 3.4% and Claude at 4.6%. On B2B SaaS the gap is wider: Perplexity hits 6.1%, ChatGPT 4.2%, Claude 5.8%. The reason is selection. Perplexity is positioned and used as a research engine with citations front and center, so its average click is a deeper-funnel click. Claude's per-visit conversion sits between the two because its users tend to be technical and patient but smaller in number. ChatGPT's conversion is lower because its sessions span everything from coding to recipes to therapy, diluting the commercial share of its outbound clicks.

Why does Claude have such long sessions on my site?

Two reasons that compound. First, Claude's user base over-indexes on developers, researchers, and analysts who read carefully and check sources before buying. Second, Claude tends to send users to long-form sources and documentation rather than quick comparison pages, so the page they land on is one they actually intend to read. In the cohort, the median Claude session is 8 minutes 3 seconds and 5.4 pages per session, both the highest of the three engines. Bounce rate on Claude traffic is 24%, the lowest. That engagement signal does not translate one-to-one into conversion because the patient-research style takes longer to convert, but it does correlate with the lowest 30-day refund rate (3.1%) and the highest 90-day customer lifetime value among the three.

How does traffic share split between ChatGPT, Perplexity, and Claude?

Across the Attrifast 200-site cohort over the Q1-Q2 2026 window, ChatGPT delivered 45% of all AI-sourced sessions, Perplexity 25%, Claude 12%, Google AI Overviews 11%, Gemini 5%, and Copilot 2%. ChatGPT is the dominant volume engine by a wide margin and that lead has compounded slowly over the last six months as the chat.openai.com search experience absorbed more queries. Perplexity has grown the fastest off a smaller base. Claude's share is small but its session quality is the highest by every engagement metric we track, so its 12% share punches above its weight in revenue contribution per visit.

What is the bounce rate for each AI engine?

In the 200-site cohort, ChatGPT traffic shows a 38% bounce rate, Perplexity 31%, and Claude 24%. All three are below the cohort median bounce rate for Google organic traffic, which sits at 49% on the same landing pages. The bounce ordering reflects how each engine pre-qualifies its clicks. Claude users land deeper in the funnel and read more before deciding, so they bounce least. Perplexity users have already seen a sourced comparison and are clicking to verify, so they bounce a little more. ChatGPT users span a wider intent spectrum and bounce more often because some of them clicked from a quick conversational answer rather than a researched citation.

Which AI engine sends the highest revenue per visitor?

Perplexity, at $2.14 RPV in the cohort, against ChatGPT at $1.18 and Claude at $1.97. The RPV ordering is not the same as the conversion-rate ordering because first-purchase value differs by engine too. Perplexity buyers spend $41 per first transaction on average, Claude buyers spend $43, and ChatGPT buyers spend $35. Claude users buy slightly bigger initial plans because they over-index on technical roles that pick a bigger tier on day one. The three-way ranking is more interesting than any one number because it shows you that the highest-converting engine is not always the one that drives the most dollars per click.

How does query intent differ across ChatGPT, Perplexity, and Claude?

The three engines route very different intent mixes to your site. In the cohort, ChatGPT sessions break down as 32% comparison, 28% informational, 24% how-to, 12% commercial, 4% navigational. Perplexity is more research-shaped: 41% comparison, 23% informational, 21% how-to, 13% commercial, 2% navigational. Claude is the most informational and how-to dominant: 22% comparison, 35% informational, 28% how-to, 11% commercial, 4% navigational. The practical reading is that Perplexity is the comparison engine, Claude is the how-to and documentation engine, and ChatGPT is a balanced mix with the broadest commercial reach because of sheer volume.

Is Claude worth optimizing for if it sends only 12% of AI traffic?

Yes for some businesses, no for others. Claude's 12% volume share is meaningfully smaller than ChatGPT's 45%, but Claude buyers in the cohort show the longest sessions (8 minutes 3 seconds), the lowest bounce rate (24%), the highest 90-day CLV ($112 versus $89 for ChatGPT and $104 for Perplexity), and the lowest refund rate (3.1%). If you sell to developers, researchers, technical teams, or anyone who reads deeply before buying, Claude punches above its share. If you sell consumer products or anything that runs on impulse, the volume gap with ChatGPT matters more and Claude is a lower-priority channel. Measure your own per-engine CLV before deciding.

Which engine is best for SaaS, DTC, and B2B verticals?

In the cohort, the three engines line up cleanly by vertical. B2B SaaS conversion goes to Perplexity (6.1%) ahead of Claude (5.8%) and ChatGPT (4.2%). DTC ecommerce flips: ChatGPT (2.8%) edges Perplexity (2.6%) and Claude (1.9%) because ecommerce runs on impulse and reach, and Claude users do not impulse-buy. Services and agency conversion goes to Perplexity (4.7%) with Claude close behind (4.3%) and ChatGPT third (3.1%). The cleanest summary: Perplexity for considered B2B, ChatGPT for B2C volume, Claude for technical SaaS and developer tools. Verify the order on your own data because vertical alone does not determine the result.

How much should I trust these numbers from the Attrifast cohort?

Trust them as a directional 2026 snapshot from a self-selected SMB cohort, not as industry-wide truth. The 200 sites are bootstrapped, Stripe-native SMBs that opted into Attrifast often because they suspected un-attributed AI traffic. That likely inflates AI share figures versus a random sample. Every number ties to a Stripe payment event joined server-side at the session level, not vendor-reported impressions, which is the strongest data quality you can get for SMB traffic. The methodology section above walks the boundaries. Re-measure quarterly on your own site because the engine mix is shifting fast and any one snapshot ages within months.

Why does GA4 fail to show ChatGPT vs Perplexity vs Claude correctly?

Three failure modes stack. First, all three clients strip the Referer header on most outbound clicks, so 65-82% of ChatGPT visits, 60-75% of Perplexity visits, and roughly 70-80% of Claude visits land in GA4's Direct or (none) bucket. Second, even a custom channel group regex only catches the minority of clicks that arrive with a usable referer. Third, GA4 cannot join the surviving sessions to a Stripe payment, so you cannot compute conversion rate, RPV, or CLV per engine inside GA4. To compare the three honestly you need server-side first-party attribution plus a Stripe revenue join, which is the architecture this benchmark is built on.

Has the ChatGPT vs Perplexity vs Claude split changed over the last six months?

Yes, slowly. From December 2025 to May 2026, ChatGPT's share of AI sessions in the cohort drifted from 51% down to 45% as Perplexity and Claude both grew faster off smaller bases. Perplexity climbed from 18% to 25%. Claude climbed from 7% to 12%. The conversion-rate gap between the engines narrowed by about 0.4 percentage points over the same window as ChatGPT added more search-style citations and Claude broadened beyond pure technical use. None of these changes is dramatic in any single month, but compounded over six months the engine mix you measure today differs noticeably from the one you measured at the start of the year.

Should I optimize content differently for each engine?

Mostly the same content works across the three, with marginal differences in emphasis. ChatGPT rewards conversational coverage and brand presence in training data, so broad topical coverage helps. Perplexity rewards fresh, well-sourced pages it can cite in real time, so source quality and recency matter more. Claude rewards long-form depth and clean documentation structure, so detailed how-to pages and reference material punch above their weight. About 70% of your content investment hits all three engines because the underlying signals (clear answers, structured headings, citable claims) overlap. Use the remaining 30% to tilt toward your highest-CLV engine measured on your own site.

How do I measure per-engine traffic and revenue on my own site?

Three layers. First, detect the engine server-side by matching the referer against a known list of AI domains (chatgpt.com, chat.openai.com, perplexity.ai, claude.ai, gemini.google.com) and falling back to behavioral inference on no-referer deep-page entries. Second, persist a first-party session row scoped to your own domain so the session survives the stripped referer and third-party cookie restrictions. Third, join that session to a Stripe checkout.session.completed webhook via metadata so each payment carries its source engine. Attrifast ships this as a 4kb cookieless script plus a Stripe OAuth connection for $29 per month with a 5-day free trial.