How Granola Grew AI Search Users 4,600%: The AEO Playbook Explained

Last Updated: 29 October 2026

Key takeaways

• Granola, an AI meeting notes app valued at $1.5 billion, grew monthly activated users from AI search by 4,600% in 7 months, according to a case study by Discovered Labs, the agency that ran the program.
• The system starts from buyer decisions, not keywords: every buying question is converted into both a Google keyword and an AI prompt, then measured.
• Every lost AI prompt is diagnosed before any content is created: missing page, weak page, crawl problem or third-party source problem each get a different fix.
• Third-party sources AI already trusts are worked with the Own, Claim, Earn, Contest framework instead of generic link building.
• Google and AI visibility are measured separately, because pages can rank on Google while staying invisible to ChatGPT.
• Alongside the AI gains, organic sessions grew 134%, making this one of the few documented AEO programs tied to a conventional SEO lift as well.

Granola, the AI meeting notes company valued at $1.5 billion, grew monthly activated users from AI search by 4,600% in 7 months. According to the Discovered Labs case study published by the agency that ran the program, non-branded AI citation rate rose 558%, non-branded mention rate rose 739%, and organic sessions grew 134% over the same period. AI search is now one of Granola's largest acquisition channels, and the playbook behind it is fully documented. This post breaks down exactly how the system works and how to run it.

What is AEO and why did Granola treat it as its own channel?

AEO, or answer engine optimization, is the practice of making a brand visible inside AI answers on ChatGPT, Perplexity, Gemini and Google AI Overviews. Granola treated AEO as a separate acquisition channel with its own baseline, its own prompt list and its own metrics, because SEO performance does not carry over. In the Granola program, pages that improved on Google sometimes barely moved in ChatGPT, and AI mentions sometimes rose while organic traffic stayed flat. AI search visibility is measured in brand mentions, citation rates and, most importantly, activated users arriving from AI-referred traffic.

According to Discovered Labs, a keyword export could show what ranked in Google but could not capture the richer questions buyers ask AI, or show whether a gap belonged to content, technical SEO or a relevant third-party source. That is why the program started with one view of demand across both Google and AI search before deciding where to act.

Step 1: Start from buyer decisions, not keywords

Granola began by mapping the questions people ask before choosing an AI meeting tool, then turned each of those decisions into both Google keywords and AI prompts. The prompt list covered scenarios, constraints and comparisons, such as which tool suits a specific workflow, how tools compare head to head, and whether a tool is worth paying for. Each prompt was then run through the AI engines and the results recorded: mentioned, cited, ranked or absent. The company and its buyers defined the market, and search tools only validated and expanded it.

Step 2: Diagnose every lost prompt before creating anything

Every prompt Granola lost was diagnosed before any content was created. Four failure modes get four different fixes: a missing page gets created, a weak page gets improved rather than duplicated, a crawl or indexing problem gets a technical fix, and a prompt where AI answers from third-party sources gets an off-page play. Keeping weak pages separate from missing pages changes the entire content strategy: new pages are only created when a real gap exists, which stops the page-per-keyword-variation sprawl that bloats most content programs.

Step 3: Prioritize technical SEO by business impact

Technical SEO work was prioritized by whether an issue affected pages tied to real buyer demand, not by tool severity scores. A crawlability or indexing problem on a page connected to an important buying question was fixed before any site-wide warning cleanup. According to Discovered Labs, technical audits become useful once the team knows what matters, makes the change and rechecks the affected page, including whether answer engines can access it.

Step 4: Work the sources AI already trusts with Own, Claim, Earn, Contest

For prompts where AI answered from third parties, Granola studied the exact Reddit threads, review sites, articles and directories that were showing up instead. Each source was checked for whether the company was absent, inaccurate, outdated, incomplete or already represented well, then classified with the Own, Claim, Earn, Contest framework. OWN covers pages Granola controls directly. CLAIM covers profiles and listings it can take ownership of. EARN covers editorial mentions, reviews and community discussions it has to deserve. CONTEST covers outdated or incorrect information that needs correcting. Different source, different action.

Reddit became a deliberate part of this. According to the case study, Granola moved its community work from broad B2B communities to communities much closer to the product and buyer in April, then began link building in June. The same buyer questions tracked in AI search decided which conversations were worth joining. "What surprised me about working with Discovered is that you have your own dedicated tool, which makes it very easy to track what's actually happening with the numbers and the effect of the different tests we're running," says Julien, Growth Marketing at Granola.

Step 5: Measure Google and AI separately, then scale what moves users

Google and AI search were tracked as two independent systems with separate baselines, set fresh whenever the scope changed. After each change the same questions were rerun: measure the gap, diagnose why it exists, make one specific change, rerun the prompts, then track whether visibility and activated users actually moved. What moved users got scaled. What did not got dropped. Over 7 months this produced the headline results: a 4,600% increase in monthly activated users from AI search, a 558% increase in non-branded AI citation rate, a 739% increase in non-branded mention rate, and a 134% increase in organic sessions.

One honest caveat on the numbers: they are self-reported by the agency, and no baseline is disclosed, so 4,600% may come from a small starting base. The 134% organic sessions lift is the most conservative figure in the set, and even that alone would justify the program for most businesses.

How to run this playbook yourself

The system is deliberately simple. List the questions buyers ask before choosing your category. Run them as prompts through ChatGPT, Perplexity and Gemini, and record mentioned, cited or absent. Diagnose every loss as missing page, weak page, crawl issue or third-party source problem. Apply the one matching fix. Rerun the same prompts monthly and scale only what moves activated users, not vanity mentions.

In our own AEO benchmarking at Flowtivity, we run exactly this kind of prompt panel across AI engines and Google for clients, and the most common first finding is the one Granola also started with: the brand is absent on high-intent comparison prompts while competitors are represented in the sources AI already trusts. Fixing that is a diagnosis problem before it is a content problem.