What an AI-Sourced Lead Is Actually Worth (and Why the Studies Disagree)

Ask five sources what an AI-sourced lead is worth and you get five incompatible answers. One study reports AI visitors converting at 14.2 percent against 2.8 percent for Google organic. Another reports AI-referred visitors converting to leads at 0.26 percent against 0.92 percent for organic, which is a 3.5x disadvantage in the opposite direction. Both were published in 2026. Both are probably accurate about what they measured.

That spread is not a data quality problem. It is a definition problem, and it matters because CFOs approve budgets against a number. If you walk into a budget meeting with a 23x multiple you found in a blog post, you will get funded once and then get audited. Here is how to reconcile the published data, build your own defensible number, and use it without setting a trap for yourself.

Why the published multiples range from 0.3x to 23x

Four things vary between studies, and each one alone can move the result by an order of magnitude.

What counts as a conversion. A newsletter signup, a free-trial start, a demo request, and a closed purchase are all reported as "conversion" across these studies. Newsletter signups convert at roughly ten times the rate of demo requests on the same traffic. A study measuring signups against a study measuring demos will disagree by 10x before anyone touches the traffic source.

What organic baseline is used. Comparing AI traffic against all organic traffic includes branded organic, which converts far better than anything. Comparing against non-branded organic is the honest apples-to-apples test, and the gap narrows sharply when you do. One ecommerce dataset across 94 sites found ChatGPT at 1.81 percent versus 1.39 percent for non-branded organic, a 31 percent advantage rather than a 500 percent one.

Sample size. The comparison roundup AirOps published makes this pattern visible: the 23x figure came from a single SaaS company, the 5x figures from a few hundred, and the smallest lifts from the largest samples. Larger samples consistently report smaller lifts. That is what regression to the mean looks like when the early studies are self-selected wins.

Whether misattributed traffic is corrected. AI assistants frequently pass no referrer, so a meaningful share of AI-sourced sessions land in Direct. If a study measures the labelled AI channel only, it captures the cleanest, most obviously AI-referred sessions and undercounts the messier ones. Which way that biases the result depends on whether the clean subset converts differently from the hidden one, and almost nobody checks.

Read down the last column and the contradiction dissolves. The studies reporting huge multiples measured low-friction actions on small samples against a generous baseline. The study reporting a disadvantage measured lead creation on a large sample. Nothing here is fraudulent. It is just five different questions with five different answers.

The four numbers that decide what a lead is worth to you

The multiple is the wrong output. What a finance team can act on is revenue per thousand AI-sourced sessions, and it takes four inputs:

1. Session-to-lead rate. Of AI-sourced sessions, what percentage create an identifiable lead record? Use your own definition of lead and keep it constant across channels.

2. Lead-to-qualified rate. What percentage of those leads pass your sales-qualified bar? This is where AI-sourced leads either prove out or fall apart, and it is the number most teams never break out by channel.

3. Qualified-to-closed-won rate. Standard win rate, segmented by source.

4. Average contract value. Segmented by source, because AI-sourced deals do not always land in the same size band as organic ones.

Multiply the four and divide into a thousand sessions:

Revenue per 1,000 AI sessions = 1,000 x session-to-lead x lead-to-qualified x qualified-to-won x ACV

The reason this beats a conversion multiple is that it survives contact with a CFO. It expresses AI visibility in the same unit as every other channel, it makes the volume constraint visible instead of hiding it behind a percentage, and it does not require anyone to believe a number from someone else's blog.

A worked example, and where it usually breaks

Take a B2B SaaS company with a $24,000 ACV comparing two channels over one quarter.

Every AI number in that table is better on a rate basis. Revenue per thousand sessions is roughly 5x organic. And the channel still produced one ninth of the revenue, because the session base is 44 times smaller.

Both facts are true and you need to present both. Leading with the 5x makes you look like a hype merchant when someone divides total revenue. Leading with the $82,500 makes the channel look like a rounding error. The correct framing is that AI-sourced sessions are the most valuable sessions you get and the scarcest, so the entire strategic question is volume, not conversion.

This is also the point where the example usually breaks in real data. With 900 sessions and 3 closed deals, the win rate has a confidence interval you could drive a truck through. One extra deal moves the AI win rate from 27 percent to 36 percent. Do not report a rate built on fewer than roughly 30 closed deals as a rate. Report it as a count with the caveat attached, and revisit at the end of the next quarter.

The volume ceiling nobody puts in the headline

The uncomfortable number in most of this research is the denominator. Seer Interactive's GA4 case study on how ChatGPT traffic converts reported AI traffic at 0.07 percent of organic sessions, roughly 11,000 sessions against 14 million over seven months. Those 11,000 sessions produced 1,370 conversions, which is a remarkable rate and a small absolute contribution.

Three implications follow, and they set the actual strategy:

1. Optimizing AI conversion rate is not the lever. It is already the best-converting channel you have. Squeezing it further returns almost nothing.
2. Volume is the entire game. Doubling AI-sourced sessions doubles AI-sourced revenue. That means citation share across more prompts, not better landing pages.
3. The zero-click majority is invisible to all of this. Most AI answers never produce a session at all. Every number in this article measures only the buyers who clicked through, which is the minority. The brand impression from being named in an answer the buyer never clicks is real and does not appear in any of these tables.

That third point is why session-based ROI understates AEO and why it is still the right number to start with. It is conservative, it is defensible, and a conservative number that survives scrutiny is worth more in a budget meeting than an ambitious one that does not.

Why AI-sourced buyers behave differently

The rate advantage is not a measurement artifact once you control for definitions, and there is a mechanical reason for it. The assistant does the filtering before the click.

A buyer who lands from Google organic has read a title and a meta description. A buyer who lands from ChatGPT has read a synthesized comparison that already named your category, described what you do, contrasted you against two competitors, and matched you to their stated constraints. They arrive further along, having pre-qualified themselves against criteria you never saw.

Seer's data shows the behavioral fingerprint: ChatGPT visitors viewed 2.3 pages per session against 1.2 for Google organic. That is not a marginally more engaged visitor, that is a different stage of the buying process.

Two practical consequences:

• Your pricing page matters more than your blog on this channel. AI-sourced visitors arrive with the education step done and go looking for disqualifying details. Missing pricing, thin integration docs, and vague security pages kill these sessions specifically.
• The lead-to-qualified rate is where the value shows up, not the session-to-lead rate. If your AI-sourced leads are not qualifying at a higher rate than organic, the engines are probably describing you to the wrong audience, which is a positioning problem visible in the answer text itself.

Three biases that inflate your number, and one that deflates it

Before you present anything, correct for these.

Inflation 1: branded query contamination. A buyer who already knows you and asks ChatGPT "is OnlyAEO any good" is branded demand wearing an AI costume. Separate branded from non-branded prompts or you are measuring your existing awareness.

Inflation 2: small-sample survivorship. Three closed deals is an anecdote. Do not annualize it.

Inflation 3: last-touch credit. If a buyer found you through a webinar in March and asked ChatGPT about you in June, last-touch hands the whole deal to AI. Run first-touch and last-touch side by side and report the range rather than picking the flattering one.

Deflation: the referrer gap. A significant share of AI-sourced sessions arrive with no referrer and get bucketed as Direct, which undercounts the channel. A self-reported "how did you hear about us" field on your demo form is the cheapest correction available, and our guide to capturing AI-sourced deals with a self-reported attribution survey covers the question wording and the ordering that avoid biasing the response.

The full plumbing for keeping these fields intact from form to closed-won sits in the CRM, not in analytics. Our walkthrough on instrumenting your CRM to report AI-sourced pipeline covers the specific fields and deal-stage records that make the four-number formula computable at all.

What to do once you have the number

The value per session number changes what you fund, in a specific order.

If revenue per thousand AI sessions is 3x organic or better, the constraint is visibility, not conversion. Spend on citation share: more prompts covered, more corroborating third-party mentions, more entity clarity. The measurement side of how OnlyAEO works exists to tell you which prompts you are losing and to whom, which is the input to that spend decision.

If the number is roughly at parity with organic, engines are likely citing you for informational queries rather than buying queries. Fix the prompt targeting before you add volume, or you will scale low-intent sessions.

If the number is below organic, check your definitions first, because that result usually means branded organic is in the baseline. If the definitions hold, look at what the AI answer says about you. Being described inaccurately sends mismatched buyers, and the fix is in the source content the engines are reading.

In all three cases the cheapest first moves are mechanical. Confirm no crawler is blocked, publish a clean machine-readable index of your content with a tool like our llms.txt generator, and make sure your content is served in a form engines can ingest without executing scripts, which is what our AI Feed Engine handles at scale. None of that earns citations on its own, but every one of them removes a reason to be skipped, and they cost a fraction of a content program.

For a worked view of how this plays out on an actual program rather than a model, our FastTrackr AI case study tracks the sequence from citation gains to session gains. The lag between those two is real and is the thing most forecasts forget to include.

The broader attribution model that all of this sits inside, including the multi-touch handling, is covered in our guide to attributing pipeline and ROI from AI-driven discovery.