HubSpot lost 80% of its traffic. Why did leads grow?

An 80% traffic decline and a 20% increase in leads were two results Kipp Bodnar cited in his UNBOUND 2026 presentation. HubSpot’s chief marketing officer said the company lost 140 million visits in a year while increasing leads. These are figures for HubSpot itself, which sells a software platform for marketing, sales and customer service.

That gap changes how a website report should be read. A visit can end after reading a reference article, or become the start of a purchase. Bodnar attributes the results to a change in audience composition: the company lost many less valuable visits while continuing to attract buyer enquiries. The traffic decline itself does not explain lead growth.

Previously, HubSpot could introduce people to the company through articles: readers moved between posts, downloaded a book and gradually learned about the product. Now, in Bodnar’s explanation, some of those conversations happen in ChatGPT and Claude. Buyers receive information within an AI interface.

People also discover brands through creators and social networks. Platforms seek to retain their audiences. Getting someone to a website is harder, while AI makes adapting a message for an interested visitor cheaper. In this account, traffic and a website’s ability to turn interest into purchases change for different reasons.

Bodnar and Kieran Flanagan propose a recurring cycle of four actions: describe the brand’s criteria and voice, tailor content for the audience, distribute it, and use results for the next attempt. In HubSpot’s documentation, the stages are called Express, Tailor, Amplify and Evolve.

At UNBOUND they focused on content production and paid advertising. The framework existed before the 2026 talk: HubSpot had already described Loop in its coverage of the 2025 conference. The September keynote is therefore useful as an account of the team’s experience and techniques, rather than evidence of an entirely new method’s invention.

Flanagan starts with sameness. When companies use similar models and prompts, well-written texts can still be interchangeable. In his view, value shifts towards choosing a message that will resonate with a particular buyer.

The team builds an evolving taste profile: who the buyer is, the language they use, why they trust the company, which stories explain the product and what attracts their attention. A model reads the profile before preparing new material. Brand criteria become part of the assignment instead of remaining in a separate unused document.

In the example shown, the same page without a profile resembled a generic software-company website. With it, the page gained recognisable HubSpot characteristics. This demonstrates the approach; it does not measure a sales increase from the new design.

Bodnar connects content quality to proprietary data, customer stories and examples. These distinguish a company’s material from what any competitor can obtain with a general prompt. Frequent publication helps maintain a presence, but a stream of similar texts does not create that advantage.

Using Sabrina Ramonov as an example, he describes two directions. Start with a long video or newsletter and prepare shorter versions for different channels. Or test short ideas first and expand those that receive a response. AI helps reshape the presentation for video, a LinkedIn post or an email; copying the same text does not accomplish that task.

Flanagan and Bodnar propose retaining quality criteria through each adaptation. More published material alone does not establish how distinctive it is or how much it interests its audience.

For paid distribution, Flanagan proposes giving platform algorithms more audience-selection work and having the team develop diverse ad ideas for different buyer tasks. He reports that automated Meta campaigns in HubSpot’s test delivered 22% higher return on ad spend than manual campaigns, and associates Google Performance Max use with a 32% reduction in the cost of acquiring a business customer.

Another result he cites: moving from 10 to 20 ad ideas a month, with the same budget and platforms, increased return by 65%. The presentation does not describe the test conditions in enough detail to apply this figure to every business. His further example of growing returns at 100 ideas is an illustration rather than a published measurement. Twenty ideas remain the speaker’s reference point, rather than a mandatory standard.

The central advertising choice is what counts as a good outcome. If an algorithm receives only form-completion signals, it can attract people who leave details but do not buy. Flanagan proposes connecting advertising to data from a CRM, a system that records customers and sales, and evaluating progress towards meetings, closed deals and revenue.

According to Bodnar, HubSpot works in cycles of two to six weeks. A two-week cadence provides 26 opportunities to review results each year; a quarterly cadence provides four. That is a frequency calculation, rather than a promise that every attempt improves performance.

AI helps gather data and draft an analysis. Employees decide what failed and which hypothesis to test next. The presentation showed a review template, but its fields were not described in the spoken account. Reconstructing a finished template from the talk would be invention.

The cycle closes when an observation changes the next idea, the taste profile or the definition of a useful advertising outcome. Flanagan explicitly acknowledges that some experiments will fail. The team needs to recognise this quickly.

The presentation offers no separate formula for distinguishing a brand’s contribution from that of the last advertisement. It also leaves AI-answer search, creator partnerships and adoption by a team without its own customer dataset largely unexplored. The 20% lead increase cannot be treated as the promised effect of implementing Loop.

The most substantive change in HubSpot’s account is the connection between content, advertising and sales. Brand criteria define what to publish; the platform determines who sees it; deal data identifies the enquiries that produced revenue. The next cycle changes decisions for which results are now available. Traffic reporting becomes part of the explanation, and marketing evaluation continues through to the purchase.