HubSpot lost 80% of its blog traffic to AI answers, and revenue grew anyway

HubSpot's blog lost roughly 80% of its organic search traffic over the past year — about 140 million visits, by chief marketing officer Kipp Bodnar's own count — while the company's revenue kept climbing, he told Search Engine Journal and DesignRush in separate interviews tied to his new book, Loop, co-written with Kieran Flanagan.

What changed

HubSpot's blog once pulled in more than 20 million organic visits a month. Bodnar says that fell by roughly 80% as AI-generated answers replaced clicks, and when Google turned on AI Overviews the blog lost close to 5 million visits in a single 30-day stretch. In the separate DesignRush interview, he puts the year's toll at 140 million visits. Revenue didn't follow it down: HubSpot's full-year 2025 revenue came in at $3.13 billion, up 19% as reported, matching the "nearly 20%" growth Bodnar cited for the year.

Why the two numbers split

Bodnar's explanation is that most of what disappeared was traffic that was never going to buy software. He told SEJ much of the loss came from low-intent posts — his example was "famous quotes" roundups — that drove pageviews but not demos. What did keep arriving, especially from AI answers, converted at roughly three times the rate of a traditional search visitor.

That cushion took years, not months, to build. HubSpot bought the newsletter The Hustle in 2021, started a creator program in 2022, and built out a podcast network and a YouTube network alongside the blog, then pruned thousands of older pages that weren't reaching buyers. About 90% of its content-sourced leads now come from channels that barely existed in its mix five years ago; the blog itself is down to roughly 10%. Its rebuilt email program, now built around AI-personalized sends, is posting an 82% conversion lift on the AI-written variant and a 50% average lift across the wider program. Qualified leads traced to AI answers grew 1,850% from Q1 2025 to Q1 2026, and its chat tool was generating more than 10,000 sales meetings a quarter by Q4. None of that replaced the blog overnight — Bodnar frames it as a multi-year rebuild that started well before the current wave of AI Overviews, not a reaction to it.

"We had tracked traffic as the scoreboard for 15 years, and it said we were failing," Bodnar told SEJ.

One thing neither interview addressed directly: HubSpot cut about 660 jobs, roughly 7% of staff, on Oct. 6. Nothing ties that round to the traffic story, and Bodnar wasn't asked about it on the record — worth knowing the growth story arrived alongside a layoff, not instead of one.

Who this affects, and what to track instead

This lands hardest on any software marketing team that still reports organic sessions as its headline content metric, and treats a falling line on that chart as proof the content function is failing. Bodnar's own replacement swaps that single number for three: revenue per visit, AI share of voice, and content-influenced pipeline. HubSpot also built a "Taste Profile" of its customers' language and a "Buyer Sim" it uses to test content before publishing, refreshing both quarterly rather than treating buyer research as a one-time exercise.

• Revenue per visit, not total visits, as the scoreboard metric for content
• AI share of voice and citation tracking, logged separately from organic rankings
• Content-influenced pipeline, so a page gets credit even when a reader never clicks through from an AI answer
• Copy written for the 40-to-100-word conversational prompt a buyer types into an AI tool, not the three-word keyword

He also described testing in six-week sprints — a thesis, a short window, and a quick stop if no signal shows — rather than letting a channel run on inertia: "You want to have a clear thesis and test it in a very short period of time to see if there's any sparks," he told DesignRush. The same discipline applies to deciding what to kill: Bodnar warns against spreading a team across every AI tactic at once, favoring a small number of focused bets that get real budget over a scattered list that gets none.