/

/

AI & Automation

Inside 12 Months of AI Search Experiments

What HubSpot learned from a year of testing AI search tactics instead of following AEO hype.

author:

AFAja Frost

Aja Frost

Director of Growth Marketing @ Mercury, former Sr. Director Growth @ HubSpot

A question to take with you

Which AI search assumption could you test this month instead of treating it as a best practice?

When HubSpot’s blog traffic fell, Aja Frost created a small experimental pod to test how AI search actually works. The team shipped weekly experiments, measured the results, and used failures to separate evidence from AEO hype.


Key takeaways


  • The team’s first llms.txt experiment produced no measurable impact, reinforcing the need to test popular advice independently.

  • HubSpot published 141 highly specific industry and use-case pages, then improved them through extensive human QA.

  • Citations arrived after crawls. Eventually, 92% of those pages were cited and visibility increased by 49%.

  • Dedicated pricing articles improved AI pricing accuracy for five of six products; correcting third-party sources closed another gap.

  • Different content can serve humans, bots, or both. The intended audience should shape the format.

Final Thoughts

Winning AI search is an experimentation discipline, not a checklist. Start with a clear hypothesis, watch how models crawl and cite information, and scale only what survives real measurement.

A short editorial introduction from pipeline.here. The full piece lives with its original author.

Inside 12 Months of AI Search Experiments

What HubSpot learned from a year of testing AI search tactics instead of following AEO hype.

author:

AFAja Frost

Aja Frost

Director of Growth Marketing @ Mercury, former Sr. Director Growth @ HubSpot

A question to take with you

Which AI search assumption could you test this month instead of treating it as a best practice?

When HubSpot’s blog traffic fell, Aja Frost created a small experimental pod to test how AI search actually works. The team shipped weekly experiments, measured the results, and used failures to separate evidence from AEO hype.


Key takeaways


  • The team’s first llms.txt experiment produced no measurable impact, reinforcing the need to test popular advice independently.

  • HubSpot published 141 highly specific industry and use-case pages, then improved them through extensive human QA.

  • Citations arrived after crawls. Eventually, 92% of those pages were cited and visibility increased by 49%.

  • Dedicated pricing articles improved AI pricing accuracy for five of six products; correcting third-party sources closed another gap.

  • Different content can serve humans, bots, or both. The intended audience should shape the format.

Final Thoughts

Winning AI search is an experimentation discipline, not a checklist. Start with a clear hypothesis, watch how models crawl and cite information, and scale only what survives real measurement.

A short editorial introduction from pipeline.here. The full piece lives with its original author.

Inside 12 Months of AI Search Experiments

What HubSpot learned from a year of testing AI search tactics instead of following AEO hype.

author:

AFAja Frost

Aja Frost

Director of Growth Marketing @ Mercury, former Sr. Director Growth @ HubSpot

A question to take with you

Which AI search assumption could you test this month instead of treating it as a best practice?

When HubSpot’s blog traffic fell, Aja Frost created a small experimental pod to test how AI search actually works. The team shipped weekly experiments, measured the results, and used failures to separate evidence from AEO hype.


Key takeaways


  • The team’s first llms.txt experiment produced no measurable impact, reinforcing the need to test popular advice independently.

  • HubSpot published 141 highly specific industry and use-case pages, then improved them through extensive human QA.

  • Citations arrived after crawls. Eventually, 92% of those pages were cited and visibility increased by 49%.

  • Dedicated pricing articles improved AI pricing accuracy for five of six products; correcting third-party sources closed another gap.

  • Different content can serve humans, bots, or both. The intended audience should shape the format.

Final Thoughts

Winning AI search is an experimentation discipline, not a checklist. Start with a clear hypothesis, watch how models crawl and cite information, and scale only what survives real measurement.

A short editorial introduction from pipeline.here. The full piece lives with its original author.

from Aisdr

Meet Ami.

Explore a different way to plan and run your outbound.

from Aisdr

Meet Ami.

Explore a different way to plan and run your outbound.

Ideas worth your time

Ideas worth your time

Ideas worth your time

A place to figure out where your next pipeline comes from

A place to figure out where your next pipeline comes from