Webflow · SaaS · run by Graphite
Eight percent of sign-ups, from a channel nobody had a line item for
TIER A
The situation
A large website builder with a mature SEO programme already running. LLM referrals were arriving but sat unattributed, folded into direct and branded traffic, so nobody could argue for budget against them.
What was actually done
Most of the lift came from SEO work already underway — use-case landing pages, templates, "how to build a website for X" pages — which served AEO without extra effort. The genuinely AEO-specific work was off-site: real named employees answering properly in existing Reddit threads, and existing YouTube output tuned toward answer-engine intent.
The outcome
Roughly 8% of sign-ups traced to LLM channels, making it one of their stronger channels though not the largest. Conversion from that traffic ran several times higher than from Google search.
Why the revenue matters more than the traffic
The conversion multiple is the whole argument. A channel delivering a fraction of the sessions at four to six times the conversion rate is not a small channel — it is a small traffic channel with disproportionate revenue attached. On a business doing a few million a year through self-serve, eight percent of sign-ups converting at that rate is the difference between a line item and a growth lever. That asymmetry is why the channel gets underfunded: the traffic report makes it look trivial and the revenue report does not.
Reported by Ethan Smith, CEO of Graphite, across a Lenny's Podcast appearance and later follow-ups. The conversion multiple is stated as 6× in one telling and 4× in another — we cite the range rather than pick the flattering end. Single-client data, not an industry benchmark. Smith has also said publicly he does not have full visibility into how Webflow's own data team validated the attribution.
HubSpot · CRM software · in-house
A 433% citation increase, and a demand number that needs reading carefully
TIER A
The situation
A company that had built its entire demand engine on Google search recognising, several years early, that the dependency was a structural risk rather than an advantage.
What was actually done
A deliberate shift away from search-dependent educational content: acquiring The Hustle in 2021, building a third-party creator network in 2022, launching multiple YouTube properties, then layering a formal AEO programme on top of that media base.
The outcome
Highest share of voice in their competitive category. Citations up 433%. Demand up by nearly 2,000%. Human-first content now drives the majority of demand, at an estimated 35–40 million engagements a month, with educational content down to roughly 28% of the total.
Why the revenue matters more than the traffic
The instructive part is not the percentage — it is that the AEO programme was layered onto four years of owned-media investment. The citations had somewhere to come from. Read as "run an AEO programme, get 433%," the number will mislead you; read as "the brands that already appear in many independent places convert citation work into demand fastest," it is the most useful figure on this page. It also reframes the spend: media investment that looked like brand marketing turned out to be citation infrastructure.
Stated by Asia Forest, Senior Director of Global Growth and Paid Advertising at HubSpot, in a conference talk. She explicitly flagged that certain figures shown in her tool walkthroughs were illustrative, while presenting these three top-line results as real verified company results. "Demand" was not defined granularly in the talk, which is why we do not treat the 2,000% figure as revenue.
HubSpot · technical fix · in-house
The pricing page that Google could read and ChatGPT could not
TIER A
The situation
Pricing lived on a page rendered client-side in JavaScript. Google processes that fine. ChatGPT and Perplexity do not render JavaScript at all — so AI services were describing HubSpot's pricing from third-party sources instead.
What was actually done
Rather than re-architecting the page, they published a series of separate pricing posts carrying the same feature and price detail in plain crawlable form.
The outcome
Pricing accuracy across AI services improved immediately and significantly.
Why the revenue matters more than the traffic
This one moves no traffic metric at all, which is exactly why it gets skipped. If an answer engine quotes your pricing wrong — from a competitor's comparison page, or a two-year-old review — the deal is lost silently, before anyone reaches your site to be counted. For a company with meaningful contract values, a handful of misinformed buyers a month is real money leaking through a rendering setting. Publishing pricing you control is cheaper than any campaign.
Described by Asia Forest with the general technical mechanism confirmed by Mike King, founder of iPullRank, on HubSpot's Field Notes. Self-reported outcome; no magnitude figure was given.
Webflow · off-site · run by Graphite
Five comments, posted under a real name
TIER A
The situation
Reddit is disproportionately cited by answer engines, largely because community moderation already filters out the low-quality material the engines would otherwise have to filter themselves.
What was actually done
An actual employee — named publicly in the retelling as Vivian — posting under her own identity, disclosing her employer, and contributing one genuinely useful answer to a thread that already existed. No new threads seeded, no second accounts.
The outcome
As few as five well-placed comments were described as potentially sufficient. The alternative approach — networks of accounts upvoting their own posts — gets detected, banned and removed.
Why the revenue matters more than the traffic
Five comments is roughly an afternoon of one employee's time. Set against a channel where conversion runs several times higher than search, this is the cheapest cost-per-acquisition in the entire discipline — and it requires no domain authority, which is why a company two years old can win a citation a market leader has not claimed. Any agency quoting you a retainer for Reddit "management" should explain what the other days are for.
Reported by Ethan Smith, Graphite. Independently corroborated by Rebecca Busk of Kvalific, who describes the same pattern and states that promotional-only posting failed while genuine problem-solving built authority. Note this directly contradicts other agencies currently selling undisclosed-account seeding.
Multiple software clients · Graphite portfolio
Double-digit pipeline influence, invisible in the analytics
TIER B
The situation
Most AI-driven exposure produces no trackable click. Buyers open a new tab, search the brand name, and arrive as direct or branded traffic. Last-touch attribution therefore credits the wrong channel almost every time.
What was actually done
Self-reported attribution added at the point of conversion — a "how did you hear about us" field on sign-up flows, demo calls and sales calls — combined with prompt-level visibility tracking.
The outcome
Once indirect attribution was counted, AI search was described as influencing double-digit percentages of pipeline and revenue for some software clients, placing it among the top three to five channels — though generally not the largest.
Why the revenue matters more than the traffic
This is the single most consequential finding for a company at a few million in revenue, because it is the difference between an internal argument you lose and one you win. Without self-reported attribution the channel reports near zero and gets defunded. With it, the same channel reports as a top-five revenue contributor. Nothing about the underlying performance changed — only whether anyone could see it. The measurement is not an accounting nicety; it is the budget.
Described by Graphite's co-founder. Practitioner estimate across a client portfolio, not an audited figure or a defined sample.
Cross-portfolio · Profound platform data
The traffic is small. The intent is not.
TIER B
The situation
Answer-engine referrals arrive after the buyer has already had their question answered and their options narrowed — a very different visitor from someone three links into a search result.
What was actually done
Conversion behaviour tracked across a few hundred customers on a platform that ingests tens of millions of answer-engine queries a month.
The outcome
Conversion rates as high as 20–30% reported from ChatGPT referrals where conversion happens directly on site. Perplexity's click-through was reported at roughly six to ten times ChatGPT's, despite far smaller volume.
Why the revenue matters more than the traffic
Judge this channel on sessions and it looks like a rounding error worth ignoring. Judge it on revenue per session and it can outperform everything else you run. That is also the trap in the Perplexity figure: a platform with a fraction of the volume can be worth disproportionate attention precisely because its users click. Volume and value have come apart, and the reporting most teams have built assumes they have not.
Reported by an AI strategist at Profound. The 20–30% figure was explicitly qualified as a high-end case heard more than five and fewer than ten times across a few hundred customers — not a typical result. We repeat the qualifier because it is the part that usually gets dropped.
Cross-web study · Graphite research
The internet is mostly AI-written. The citations are not.
TIER B
The situation
Analysis of a large Common Crawl sample suggested AI-generated content now outnumbers human-written content across the open web.
What was actually done
The same detection method applied specifically to content appearing in Google results and ChatGPT citations, validated against a pre-ChatGPT-era sample to establish a false-positive baseline.
The outcome
Only around 10–12% of content that actually gets cited or ranked was AI-generated. The correlation runs against fully automated writing, not for it.
Why the revenue matters more than the traffic
The volume play is the most expensive mistake available in this field right now, because it fails quietly. You spend a year producing hundreds of pages, the traffic report shows movement from indexation, and the citation share never arrives. Meanwhile roughly one landing page in twenty drives about 85% of a site's traffic — so the correct move was almost always fewer pages, better made. AI-assisted with a human editing is fine and increasingly standard; fully automated at volume is a budget line with no revenue attached.
Graphite research, using a third-party AI detector with an approximate 8% false-positive rate established by testing against a 100,000-URL pre-ChatGPT sample. Detector-based methodology, so treat the precise percentage as directional.
