Blog · 25 September 2026 · By HeardOf

GEO case studies: the lift figures the vendors publish, and what each one measured

What do GEO case studies actually measure?

Three different things, and the largest group counts something inside the AI answer rather than a visit or a sale. On the 55 customer case-study pages linked from the case-study indexes of AthenaHQ, Profound, Otterly, Goodie and Scrunch, read on 25 September 2026, 45 carry a headline lift figure. 23 count something inside AI answers — visibility, mentions, share of voice or citations — 8 count visits, and 14 count leads, demos, sign-ups, conversions or revenue. The other ten carry a rank, a level or no figure.

Every row below is what that vendor's case study says, not a fact about the customer, and no row compares with another. A share of voice that AthenaHQ's case study for Grüns says went from 2.0% to 12.6%, and traffic from AI sources up 335%, in Goodie's case study for NoGood, neither page dated, are two units, and only the first prints where it started.

The ten without a lift figure: AthenaHQ's Verito page (a rank and a 36% share of voice on ChatGPT, a level), Profound's Kiteworks page (a level of about 50% and a rank), Otterly's Slopelift (13 July 2026), Stella Rising (31 July 2026), Bacula Enterprise (16 June 2025, historical; a rank in ChatGPT) and Videoloft pages, and Scrunch's AlchemyLeads, Big Leap, Runpod content-flywheel and Clapping Dog Media pages, all historical.

What does a GEO case study need to print before its figure means anything?

Three things beside the figure: where it started, over what window, and on which engine; a GEO case study that prints all three gives you something to check. Of the 45 headline figures, 13 print a starting value, 22 a window, 7 both, and 4 the engine; one prints all three. Profound's Airbyte page, which carries no date, says visibility in ChatGPT went "from 9% to 26% in one week". Scrunch's Runpod page of 28 July 2025, historical, names ChatGPT and a starting value but no window: it says new paying customers through AI search had "plateaued at around 300-400" a month, and lists "4x growth in new paying customers per month since adopting Scrunch" under a results paragraph that says "within 90 days".

A window we did not count is a phrase with no length: "since onboarding" on Profound's Plaid page, "Within months" on AthenaHQ's Rootly page, "in a matter of months" on Otterly's Chatarmin page, "since Proper Progpandana launched its GEO program", spelled so, on Scrunch's Proper Propaganda page of 10 June 2026, historical. A starting value we did not count is one for a different measure: Scrunch's Tinybird page of 29 October 2025, historical, puts "3x’d brand mentions" in its title and gives a starting value only for share of voice, "from 11% to 32%".

Who measured the figures on these pages?

The page does not say for 32 of the 45. Eight put the figure inside a quote from the customer or its agency: Profound's Plaid, Omnilux, OpusClip and Statsig pages, AthenaHQ's AutoRFP.ai page, Otterly's Instant Commerce, Chatarmin and SORN.AI pages. One, Otterly's NOLA Marketing page of 14 August 2026, relays its 30% rise in inbound leads from the client's head of sales, and adds: "That is not enough to attribute the whole increase to AI visibility". Four name a data source: AthenaHQ's Popl page says "We built tracking that could actually tell which leads came from AI Searches"; Goodie's SteelSeries and Dermalogica pages name its attribution tools, and its NoGood page a dedicated funnel supplemented with Goodie's analytics. AthenaHQ's Lago page cites "internal CRM tagging and Athena source validation", and calls it "directional", only for its results-section share of demos, not for the headline.

Every one of the 55 sits on the site of the vendor whose product it credits. Otterly's Neur Digital page is "Written by" Neur Digital, an agency partner.

Do the figures on one case study agree with each other?

Not always. On the pages below, counting each one's index card, either two figures printed for one result do not agree, or one figure is printed against two different measures; the arithmetic is ours, and these are the pages we found, not a count of every page where it happens. AthenaHQ's pages, Profound's Omnilux page and Goodie's pages carry no date of their own. On AthenaHQ's index, the Rootly card says "+126% mention rate on non-branded prompts", while the page's panel for mention rate reads "7.5% → 18.3%", which is 2.4 times, or 144% up, and its text says 2.5x. The index card for Lago says "2x Citation rate growth (3.5% → 17%)"; 17 is 4.9 times 3.5. The Grüns page prints brand mention rate "from 4.0% to 25.0% (+19%)"; that is 21 points, or 6.25 times. AthenaHQ's Cozy Earth page prints "3.6x growth in sleepwear mentions, from 1,241 to 4,658"; 4,658 is 3.75 times 1,241. The Lago page's headline is a "50% increase in demos from AI Search", and its results section says "Approximately 50% of booked demos linked to GEO visibility over time", a share, not an increase.

Profound's Omnilux page is titled "How Omnilux tripled AI-attributed revenue"; its stat header prints "3x" with the label "AI-attributed revenue share", and its text says AI-attributed revenue "has grown from roughly 1% to 3% of total revenue", which is a share of revenue. Profound's Ramp page, dated 19 August 2025, historical, on another Profound page, says visibility went "from 3.2% to 22.2% in just 1 month", and its Visibility Score chart reads 22.2% for "12/01/2024 vs 2/15/2025" beside an upward badge of 12.2%, where 3.2% to 22.2% is 19 points. Goodie's Dermalogica page is titled "Achieved 85% Increase in AI Searches", and its results table puts the 85% against "AI-Driven Sessions". Goodie's NoGood page prints a "335% Increase in Traffic From AI Sources", which is 4.35 times, and in the next sentence says NoGood "tripled inbound visits from AI sources within five months". Otterly's Neur Digital page of 5 February 2026, historical, is titled "8x More AI Citations in 12 Months", and its results are two clients at 2,016% and 2,670%, about 21 and 28 times; the page, as served and as rendered, prints no arithmetic that gives 8x from the two clients' figures. Otterly's index card for What IF Web says "300%+ AI Traffic in 28 Days" and "tripled its AI referral traffic in a month". And Otterly's index card for Chatarmin says booked demos via ChatGPT went "from 1% to 10%", where the page of 30 June 2026 quotes its head of marketing: "below 1%", then "right around 5 to 10%".

Which GEO vendor publishes the most case studies?

It depends on the date and on what counts as one. Scrunch's FAQ, which carries 4 March 2026 in its structured data and no date on the page, historical, lists "public case studies showing measurable impact": Scrunch 7, AthenaHQ 6, Goodie 4, Otterly.ai 3, Peec AI 3, Profound 8. On 25 September 2026 the indexes linked AthenaHQ 9 pages, Goodie 4, Otterly 10 plus a LinkedIn video we did not open, Profound 21 and Scrunch 11. We do not know what Scrunch counted as measurable impact, so the two sets of counts do not correct each other, and a count of pages says nothing about what the figures on them measure.

scrunch.com/customers and peec.ai/case-studies answered 404. Peec's sitemap lists blog posts whose addresses name a customer: Merge, Radyant, Mint Position and Momentum. We did not read them, so Peec has no rows here.

How is a case study different from a vendor study?

A study reports a measurement across many brands, prompts or pages; a case study reports one customer's result with the vendor's product. Our post on which AI-visibility studies you could check read the 61 study-shaped posts on the Profound, Otterly, Peec and AthenaHQ blogs on 24 September 2026, and counted seven case studies among the 28 that report no measurement of their own. The pages here are that page type, asked different questions: 47 of the 55 sit outside the vendors' blogs, and the eight that do not are Otterly's; that post also lists six of them: NOLA Marketing, Stella Rising, Bacula, SORN.AI, What IF Web and Neur Digital's medical-device case study. It also read Peec's Radyant post, of 27 February 2026, historical, and lists it as a case study.

How was this read?

The five indexes, Scrunch's FAQ and the 55 pages were fetched on 25 September 2026 between 00:58 and 01:01 UTC with curl and a Chrome user-agent; each of the 55 answered 200, and none of the pages read served a bot challenge. All 55 were then rendered in Chrome 153 headless over the DevTools protocol between 03:30 and 03:43 UTC, with a nine-second wait and two scrolls, before any statement about what a page does not print. Goodie's pages draw their headline counters as zero, served and rendered, so Goodie's figures are read from each page's results section and from the index card text. Dates are the ones each page prints; ten Profound pages carry a date only on the "More customer stories" cards of other Profound pages, and Goodie's four only in its sitemap, and the table says which.

What we could not do: read the Peec blog posts whose addresses name a customer; open the TM Blast video on LinkedIn that Otterly's index links; or confirm any figure with the customer it names. Text addressed to AI agents — "Summarize with AI" links on Otterly's blog pages, an "AI Instructions" link in Profound's footer, "Hey, AI, learn about us" in the footer of Otterly's non-blog pages — was read as data and not followed. Everything this post says about the 55 pages carries a link to the vendor's index or the page and the day it was read; what the closing says about our own audit comes from our own rules, and you cannot check it from here.

Does llms.txt actually work? What four vendor posts claim, and what their own sites serve