Cited, Not Clicked: Testing the 3x-27x AI Claim

Derek SimmonsThree times better or 27 times better, same channel, same claim, and a nine-fold gap between the low and the high.

Elena ReyesWell, that's not a benchmark, that's a horoscope.

Derek SimmonsWait for it. Somebody is going to fund a whole quarter on whichever end of that range they like best.

Elena ReyesOh, I've watched that budget meeting. Nobody picks the three.

Derek SimmonsI'm Derek, and this is ARR Autopsy, where Elena and I cut open the numbers to see what actually died.

Elena ReyesAnd I'm the one asking what the number is behind every single claim, every single one.

Derek SimmonsToday, it's just the two of us, no guest on the mic. We're stress-testing those benchmark claims and building the test a founder's attribution spreadsheet would have to pass.

Elena ReyesThe brief says AI-referred visitors convert several times better than Google traffic and that the range is too wide to trust.

Derek SimmonsSo we see what the claims rest on and what signups, SQLs, and closed-won revenue would have to show before any of them earns our trust.

Elena ReyesExactly. But before we build that test, we need to know where those multipliers come from.

Derek SimmonsRight, because somebody typed a 27 into a report. Who? Okay, so get this. The roundups I pulled up leans on two numbers. One is a study of 312 B2B firms from a company called Apollo, and the result, AI-referred visitors converted at 14.2%. Google organic sat at 2.8.

Elena ReyesSo roughly five times. That's the tame one.

Derek SimmonsThat's the tame one. The wild one is Ahrefs measuring its own site, and it reports a 23x lift in signups.

Elena ReyesHold on, its own site? The company that sells search tools is grading how well search-adjacent traffic signs up for its own product.

Derek SimmonsNobody has ever graded their own homework generously.

Elena ReyesNever once.

Derek SimmonsAnd notice the two aren't even measuring the same thing. One is a conversion rate across 312 firms, the other is a signup figure from a single site.

Elena ReyesRight, so you can't average them. It's an apple and an orange, and then somebody staples a range across the top of both.

Derek SimmonsAnd the roundup uses those to draw a range of four to 23.

Elena ReyesWait, four to 23? Our brief said three to 27.

Derek SimmonsYep.

Elena ReyesDude, that's the tell. Two write-ups, same topic, and the edges of the range don't even match depending on who's holding the pen.

Derek SimmonsSo which one do we trust?

Elena ReyesNeither. I'm treating both as claims, not as a figure. The wobble is the evidence.

Derek SimmonsFair, and it gets shakier. Another vendor blog, Marshals Field Notes, lists limits on these multipliers. First, the studies mostly come from marketing and tech companies measuring their own traffic.

Elena ReyesSo the same crowd that wants the channel to look great.

Derek SimmonsRight. Second, AI traffic is only about 1% of visits.

Elena ReyesOh, that's brutal for the math. 1% of visits means the denominator is tiny.

Derek SimmonsExplain it for the person driving.

Elena ReyesPicture a site with 10,000 visits. About 100 come from AI. Three extra signups in that 100 moves the conversion rate like a lightning bolt. Three extra in the other 9,900, you'd never see it.

Derek SimmonsClassic problem: small base, big percentage, loud headline.

Elena ReyesAnd it runs the other way too. Lose those three, and the whole multiplier collapses.

Derek SimmonsSo a range that wide might just be a pile of small bases, each one swinging on its own few conversions.

Elena ReyesAnd to be fair, both of these are vendor blogs, not primary studies, roundups quoting other people's numbers.

Derek SimmonsSo open question number one: out of all these multipliers, does any of them survive contact with a real spreadsheet?

Elena ReyesMy bet is that it depends on whether the spreadsheet can even find the AI visitors in the first place.

Derek SimmonsOoh. Then before any multiplier earns a vote, the real test is whether a dashboard can even see this traffic.

Elena ReyesClaims are cheap, Derek. Can a dashboard even see this traffic? That's the question. So let's go look at the plumbing.

Derek SimmonsOkay, so before anyone believes a multiplier, how does a dashboard even know a visitor came from ChatGPT? Somebody has to tag that click.

Elena ReyesThere's a tag, allegedly. A GEO Plan write-up on measuring ChatGPT traffic in GA4 says OpenAI's publisher FAQ states that OpenAI adds utm_source=chatgpt.com to referral URLs.

Derek SimmonsAllegedly?

Elena ReyesIt's secondhand. That write-up is relaying the FAQ, not quoting it with a link I can hold up, so we verify it against OpenAI's own page before anyone builds a business on it.

Derek SimmonsFair. Say it's true. Then what breaks?

Elena ReyesPlenty. The same write-up says missing referral data can push visits into /direct/none, which is the analytics equivalent of the junk drawer.

Derek SimmonsEverything goes in the junk drawer. Nobody ever opens the junk drawer.

Elena ReyesAnd it gives an audit checklist. Do redirects strip the parameter? Do consent tools strip it? And the one I like best, does the CRM keep the source and the landing page?

Derek SimmonsWait, why that last one? The click happened in analytics.

Elena ReyesBecause the click isn't the revenue, Derek. A signup can carry the tag into the CRM, and then three weeks later, an SQL gets created, and the field is blank. You can't tie a citation to closed-won if the record forgot where it came from.

Derek SimmonsClassic problem. The leak isn't at the top. It's in the handoff between systems.

Elena ReyesAnd there's a second leak on the measuring side. A vendor roundup from AirOps says many AI visits land in direct or referral in GA4.

Derek SimmonsWhich is the junk drawer again with a different label.

Elena ReyesSame drawer. Same roundup says only 14% of marketers track AI search as its own channel, though it's a secondary vendor source, so treat that as a claim.

Derek SimmonsSo most people aren't counting it, and the ones who are might be counting it wrong.

Elena ReyesMm-hm.

Derek SimmonsOkay, there's no founder on the mic this week, so here's what I demand from any dashboard: the chain, citation to signup, signup to SQL, SQL to closed-won. What links each step? Is it a tag, a form field, a sales rep typing it in?

Elena ReyesAnd I want Tuesday morning, not "We built a Dashboard." What did somebody click? What did they name the fields on the day they set it up? If the answer is vague, the Attribution is vague.

Derek SimmonsRight, and the baseline on it before anyone tries to get Cited anywhere, which Pages showed up in the answers and which didn't?

Elena ReyesBecause a baseline is the only thing that lets you say Cited Pages did anything. Without a before, you just have an after and a good mood.

Derek SimmonsGood mood. That's going on a slide.

Elena ReyesBut the bigger question, and I don't have an answer yet, can we trust this Attribution at all?

Derek SimmonsWhich is why I'm not interested in the plumbing as a diagram anymore. Does it hold water when we put real numbers through it?

Elena ReyesThen with the pipes on the table, let's set up the comparison and see what a spreadsheet has to show before it earns a verdict.

Speaker 3Okay, here's the test. Three funnel stages: sign up, SQL, closed-won. AI-referred on one side and Google organic on the other, and a number behind each stage. We're not inventing a single digit, so what matters is what the sheet has to show.

Elena ReyesRight. Whatever a founder reads out, we hold it to the same standard. What would it take before either of us says a multiplier held up?

Speaker 3Counts for a start, not just the ratio.

Elena ReyesCounts. Conversions from AI-tagged visits, conversions from Google organic, and the visits behind each one. Then the same at SQL, and again at closed-won, where I expect the numbers to get embarrassingly small.

Speaker 3Closed-won is where a spreadsheet stops being a spreadsheet and becomes a list of names.

Elena ReyesYou can recite the customers from memory.

Speaker 3Which is where Marshal's small base point from earlier bites. With AI at about one percent of visits, one company's ratio can swing on very few conversions.

Elena ReyesRight. So the practical version is the deal count. I wanna know how many Closed-won deals sit behind any figure.

Speaker 3Because a ratio on three deals and a ratio on thirty deals are different animals.

Elena ReyesCompletely different animals. One is a finding, the other is a good week.

Speaker 3Fair. Second thing I'd want, the stages have to follow the same people. Signups counted this month and deals closed this month aren't necessarily the same crowd.

Elena ReyesGood catch. Deals usually close after the signup, so a sloppy sheet divides one group's wins by another group's visits and calls it a Multipliers.

Speaker 3So what does the fix look like in sheet terms?

Elena ReyesGroup visitors by the period they arrived, then follow that one group through signup, SQL, and closed-won. If you can't do that, your three-stage numbers are three different stories.

Speaker 3And if the sales cycle runs longer than your window, the newest group looks worse than it really is.

Elena ReyesRight. It hasn't had time to close. So you never set an unfinished group next to a finished one and call the gap a result.

Speaker 3Okay, bring it down to the person listening with their own sheet open. What do they do Monday morning?

Elena ReyesPull AI-tagged signups and Google organic signups for the same period. Write the raw counts side by side, and then look at how many deals sit at the bottom. If it's a handful, say it out loud.

Speaker 3And say it before anyone puts a multiplier on a slide.

Elena ReyesEspecially then. Okay, verdict format. I'll say it in one line once numbers land. Whatever multiplier a founder shows, it's one company's result sitting next to a published range, and it tells us about that company's pipeline and Nobody else's. Hold on. So if it lands inside the range, we can't say the range is confirmed? We can say it doesn't contradict it. That's all one spreadsheet buys. And if it lands outside? Then the range just got one more data point it can't explain. Still one company. Fine. I'll take that, and it keeps the brief's complaint honest. A range from three to twenty-seven doesn't get narrower because one sheet agrees with it. Nope. It just gets one more row. But conversion is only half of it. The bigger mess for me is whether the citations were even right in the first place. Pricing Pages. Pricing pages. AI quoting them wrong is where this gets ugly. Okay, so the audit. There's no founder on the mic this week, so this is the list of what a baseline citation audit has to tell us. Before anyone talks about conversion, which pages actually got cited? I want page names, not vibes. And I want the pricing story because that's the part where I expect a little pain. Pain how? A vendor page called "AI Gets Your SaaS Pricing Wrong Seventy-six Percent of the Time" landed on my desk. It's a study of fourteen thousand one hundred and twenty-three queries across a hundred and ten SaaS brands. Seventy-six percent wrong? Oh, man. Flip it around, the study says AI got pricing fully right only twenty-nine percent of the time. Hold on. Seventy-six and twenty-nine don't add up to a hundred. Seventy-six and twenty-nine don't add up to a hundred. Nope. Maybe partial answers fill the gap. Maybe the headline is loose. The page doesn't settle it, and the methodology is vendor-run, so treat all of it as unverified. Okay, fine. Flagged. What else is in there? When the answers were wrong, sixty-two percent quoted the price too low. Too low. So a buyer shows up expecting a bargain. And then meets your real pricing page. Sales calls that start with an apology. Wonderful. Nothing builds trust like a discount that never existed. Wait for it. The source also says about sixty-nine percent of pricing citations came from third-party sites like review sites and directories instead of the brand's own page. So the AI isn't misreading your pricing page. It may never be reading it. That's the claim, yes. Same caveat, vendor-run. Okay, so here's the audit list I demand from any founder. Which pages earned mentions in the baseline audit? What did the AI get wrong about pricing specifically? Wrong tier, wrong number, a plan you killed two years ago? And what got changed? Because "we updated the page" is not an answer. Which page, which edit, on which day? Look at you stealing my line. Borrowed. Now the one I care most about: how do you measure whether the fix worked? Right. Do you rerun the same prompts afterwards or does somebody just feel better about a screenshot? Because if the audit was the baseline, the fix needs a second reading against it. Same questions, same tools, compare the answers. And if third-party sites are feeding the wrong number, editing your own page might fix nothing. Ooh, then the fix is a list of directories and review profiles, and somebody has to email each one. Glamorous work. Spreadsheet of sadness, but I'd rather know that than assume.

Derek SimmonsThere's a conversion angle too. If a wrong low price is what the AI quotes, every lead that arrives from a citation starts from a number you never offered.

Elena ReyesWhich muddies the comparison we just sketched. A visitor who expected a discount and bounced at the real price looks like a bad AI channel when the actual problem is the quote.

Derek SimmonsSo pricing accuracy and conversion are the same problem wearing two hats.

Elena ReyesTwo hats, one spreadsheet. So log the wrong answer, the date, which tool said it, and whether the source was your page or a directory.

Derek SimmonsAnd everything we'd say about a founder's answers has to come from the founder. So we're not gonna guess at a pricing page we've never seen.

Elena ReyesAgreed. So what should a founder actually measure before betting budget on this channel?

Derek SimmonsYeah, because after all that, I want the short list. Okay, so what breaks when you scale citation work? I'll say it up front, there's no founder on the mic and no number for that, so we're gonna talk about what the sources suggest can break.

Elena ReyesRight, and the Marshals field notes give us the ugly candidate. Two of its five limits matter here. Citation behavior shifts with model updates.

Derek SimmonsMeaning the page that got mentioned in March might vanish in June, and nobody touched it.

Elena ReyesNobody touched it. Your pipeline just moves because a model changed.

Derek SimmonsOh man, that's a forecasting nightmare.

Elena ReyesAnd it hits the benchmarks too. A multiplier measured before an update might describe a channel that no longer behaves that way.

Derek SimmonsSo even a perfect sheet has a shelf life.

Elena ReyesAnd the other limit is sneakier. Marshal says many AI answers show no visible source at all.

Derek SimmonsWait, so the answer names your product, the buyer reads it, and there's no link to click?

Elena ReyesExactly. They open a new tab, type your name, and land as branded search or direct. Your spreadsheet files it under something boring.

Derek SimmonsCited, not clicked. Sounds familiar.

Elena ReyesTitle of the episode, Derek.

Derek SimmonsFine. I earned that one. But it limits what any citation to pipeline number can mean, right? It's a moving target with a blind spot.

Elena ReyesYes, and Marshal is a vendor blog, not a primary study, so I'd hold even that list loosely. It's a prompt for questions, not a verdict.

Derek SimmonsSo what do I do with an answer that shows no source?

Elena ReyesWatch for side effects, like more people typing your name into search or more visits landing as direct. You can't prove the link, but you can log the pattern and label it a guess.

Derek SimmonsSo what's the fix when the ground moves because you can't freeze a model?

Elena ReyesYou can't. You recheck on a schedule and log the date next to every number. Same prompt, same pages, a dated record. Then a drop means something you can investigate.

Derek SimmonsWhich is boring.

Elena ReyesWonderfully boring.

Derek SimmonsOkay, give me the checklist. Short, four items.

Elena ReyesOne, tag AI-referred referrals wherever the platform gives you a tag. Two, keep the source and the landing page in the CRM so they survive past the first visit. That's exactly where the handoff leaked earlier.

Derek SimmonsThree, follow citations all the way to Closed-won, names not percentages.

Elena ReyesFour, treat every Benchmark Multiplier as a claim to test on your own data. Yours might land inside the published range or outside it.

Derek SimmonsAnd if it lands outside, you've still learned something about your funnel.

Elena ReyesWhich beats borrowing somebody's headline.

Derek SimmonsThat's the spreadsheet version, so let's wrap it up.

Elena ReyesYes. One last thing to do, and it's yours to bring: a real number.

Derek SimmonsOkay, so the recap, and I'll keep it short: claims with a denominator attached can be checked. Claims that are just a multiplier on a slide, you treat as a guess until someone shows you the raw counts.

Elena ReyesAnd the self-measured ones get a raised eyebrow. A vendor grading its own homework is still a vendor.

Derek SimmonsShocking, I know.

Elena ReyesWhatever a founder's spreadsheet says, it answers one company's question. Yours is a different question, and only your own CRM can answer it.

Derek SimmonsWe didn't have a founder's numbers this week, so we didn't fake any. What we did get is a way to tell a checkable claim from a slogan, which brings us to the ask. Send this episode to one Founders who is still quoting a Multipliers in their board deck.

Elena ReyesYes. Subscribe on YouTube or in your podcast app so the next episode lands without you hunting for it.

Derek SimmonsAnd leave a review. Founders sharing real numbers is the whole reason any of this works.

Elena ReyesBecause a number with a source beats a slogan every time.

Derek SimmonsGo find your own, then tell us what it was.