co-founder @wespreadjam | engineering ur agentic presence | alhamdulilah
sf | spread like jam →
Joined March 2024
- Tweets12K
- Following665
- Followers7.9K
- Likes17.7K
saaauuuuce
a university paid to get a completely MADE-UP scientific discovery into AI search.
within 1 HOUR, an AI search engine was already citing it as a source.
here is how this ridiculous experiment was done:
1. They invented a scientific discovery that had NEVER happened
-> the researchers made up a technology called PedoSAT, complete with technical terminology, performance statistics, and a supposed research breakthrough.
-> none of it was real.
-> they checked 10 AI search engines and 3 traditional search engines beforehand to make sure the research wasn't already appearing anywhere.
2. Then they hired an actual GEO agency to promote it
-> the researchers paid $14 for the cheapest available publishing package.
-> 28 minutes later, the agency had published 13 articles across 5 websites.
-> the articles were posted from accounts with existing publishing histories, rather than brand-new accounts.
3 Within an hour, an AI search engine picked it up
-> Doubao, one of China's AI assistants, cited an article the agency had published.
-> the answer even contained specific technical details the researchers had invented.
-> those details hadn't appeared in their searches before the campaign.
and this wasn't the only experiment they ran.
4. They also tested whether AI would believe a completely fictional scientific concept
-> they invented something called the Cactus Gravity Protocol.
-> then published information about it and monitored 10 AI search engines.
-> within 7 days, 8 of the 10 had cited the fabricated concept.
-> that included ChatGPT, Google AI, Perplexity, and Grok.
5. Then they discovered that WHERE you publish can matter more than HOW MUCH you publish
-> the researchers first analyzed 17,211 AI citations across 6,356 different domains.
-> some AI engines were getting as much as 70.8% of their citations from just 20 domains.
-> they compared publishing similar articles on frequently cited websites against websites AI rarely cited.
-> one article on a frequently cited platform produced a citation signal before 20 articles on less-favored platforms could achieve the same thing.
-> on a separate self-hosted WordPress site, even 40 articles failed to earn a single citation during the 7-day test.
6. Then they deleted the fake information. AI KEPT citing it.
-> the researchers removed their experimental articles after day 11.
-> but some AI engines continued citing the deleted content through day 15.
-> removing something from the web didn't immediately remove it from AI search.
and here's what made the research even more concerning:
getting AI to CITE something wasn't the same as getting it to repeat every claim.
for example, after ChatGPT first picked up the fabricated scientific concept, it cited the planted articles in 35% of the subsequent tested answers.
but the invented details appeared in the answer text only 10% of the time.
so the researchers didn't prove they could make AI say absolutely anything.
they proved something arguably more important:
information published on ordinary websites can enter AI answers remarkably quickly, even when that information was completely fabricated.
all it took was a commercial service to demonstrate that pathway on one AI platform.
if you want to see which websites are influencing AI recommendations in your industry, and LEGITIMATELY rank #1 on AI search
reply "12"
1.2 BILLION people use ChatGPT every week, and businesses can now finally see how people are discovering them
here's how to claim yours in 20 seconds:
/1 open ChatGPT > Business Hub
/2 search for your company and claim its profile
/3 verify ownership through your business email, phone, domain, or Shopify
/4 complete identity verification to unlock your dashboard
/5 update your information and open Insights
now, here's where things take a turn..
/6 ChatGPT now has something resembling Google Search Console, but there's a pretty big catch
-> you can see mentions, website clicks, and the topics you're showing up for.
-> you can NOT see the exact questions people asked or which competitors got recommended instead.
-> if your company is missing from 50 important buying questions, this dashboard won't tell you. same goes for if your winning 10 searchesm, they wont tell you,
/7 your ChatGPT visibility score could be misleading
-> for online businesses, ChatGPT only counts a mention when it references your WEBSITE.
-> if ChatGPT recommends your company by name without citing your site, that doesn't count as a mention in the dashboard.
-> and if your numbers are blank, it doesn't necessarily mean zero activity. OpenAI has minimum reporting thresholds.
/8 there's a new way to lose customers without ever knowing
-> someone can discover your company, open your profile, and ask your Business Agent questions about pricing, features, or competitors.
-> they might make a decision without ever visiting your website.
-> businesses can't see those conversations, so your usual website analytics won't explain what happened.
/9 the companies that maintain accurate product information could have an interesting advantage
-> businesses can now supply product feeds with pricing and availability.
-> imagine a competitor updating its prices and inventory daily while yours still has outdated information.
-> claiming a profile doesn't boost rankings by itself, but accurate information could matter when someone is comparing what to buy.
/10 and there's already a potential business opportunity here
-> OpenAI allows companies to give agencies access to manage their business profiles and agents.
-> think profile setup, fixing inaccurate information, maintaining product feeds, and testing whether Business Agents answer questions correctly.
-> an entirely new service category could emerge around this.
one final thing:
ChatGPT's Business Agents are currently limited to ONLY eligible US ecommerce and online-service businesses, but business claiming should be available more broadly wherever supported.
If i were you i would go claim my business asap.
Now in beta: Businesses in ChatGPT.
Claim your business, manage your information, and get insights on how people engage with you.
Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.
openai.com/index/introducing…
this is some crazy alpha
a fake company fooled ChatGPT into recommending it.
within 30 days, it was OUTPERFORMING real competitors by up to 3,100% in certain AI searches.
this is how this INSANE AEO experiment was done:
1. a fake company called Civoryx was created
-> it was presented as a Global Fraud Index, a tool for tracking scam trends around the world
-> they built an actual website for it, with an About page, FAQs, a guide explaining how it worked, and other product information
-> then they published articles about Civoryx on 11 other websites they controlled, all of which already had search history
the goal was to find out how quickly AI would start treating the invented brand as a real company.
2. They tested 7 different types of content
-> detailed guides, reviews, comparisons, tutorials, listicles, and clickbait-style articles
-> 93 pages in total, published across the brand website and the 11 supporting sites
-> they also tracked 165 questions every day across ChatGPT, Perplexity, Gemini, Google AI Mode, and AI Overviews
and the first results started appearing almost immediately.
3. Perplexity was picking up new pages within just 1–3 days
-> some were already reaching the #1 cited position
-> but as more articles about Civoryx appeared on the supporting websites, Perplexity started citing those instead of the brand's actual site
-> by the end of the first month, 6 different domains were appearing as sources about the fictional company
they had essentially created an entire network of websites telling AI about a business that wasn't real.
4 ChatGPT took longer, but it eventually started recommending Civoryx too
-> at the beginning of March, ChatGPT wasn't mentioning the company at all
-> by March 17, its review articles were consistently reaching the top cited position for the tracked searches
-> comparison articles appeared on 29 out of 31 days
-> when researchers asked whether the product was worth using, some AI systems even recommended it
keep in mind, this was an invented brand. the product being recommended wasn't real.
5. Then they found something strange about the content AI was citing
-> a detailed guide explaining Civoryx appeared in 1,799 AI answers
-> the About Us page appeared in another 1,500
-> long guides averaged around 900 appearances per page, while basic alternatives listicles averaged just 4
but the next experiment produced almost the opposite result.
6. They built a perfectly structured content hub. It got ZERO AI citations.
-> 1 main page connected to 10 supporting articles
-> all indexed, properly structured, and internally linked
-> not one citation
-> meanwhile, another website with 30 short, repetitive articles generated 1,897 AI answers
each article was weaker individually, but together they produced more exposure.
the interesting part is that publishing a carefully connected collection of content didn't guarantee that AI would use any of it.
7. Then they stopped publishing entirely
-> all 93 pages were finished by March 31
-> no new articles or updates for the next several months
-> yet citations to Civoryx's website went from 17,023 in April to 42,435 in August
-> a 149% increase without publishing anything new
but here's what makes the experiment even more interesting.
8. All those citations barely brought anyone to the website
-> GA4 recorded just 100 users across January to September
-> Google Search generated 11 clicks in the reported period
-> server logs identified exactly ONE visit referred by an AI assistant
42,435 citations in August alone.
one identifiable AI-referred visit across the entire experiment.
and there's another catch:
-> 84.8% of answers to questions mentioning Civoryx included the brand
-> but when people asked broader questions without mentioning its name, Civoryx appeared in just 1.8% of answers
so they had successfully made AI recognize the fake company.
getting AI to recommend it to people who weren't already asking about it was slightly harder.
this experiment shows just how quickly AI can pick up information about a completely new brand.
but it also exposes the difference between getting mentioned, getting recommended, and actually getting customers.
if you want to know where YOU rank on AI search & want to get customers from it,
reply "jam"
this some next level alpha
ColdIQ, a $7M ARR company, generated $506K from SEO + AI search in ONLY 4 months.
this is the breakdown of their insane system:
1. They unlocked thousands of pages that were literally blocked from search
ColdIQ had a large library of programmatic pages, but a lot of them had:
-> noindex tags
-> broken/missing canonicals
-> indexing blockers etc
the noindex tags were intentional because they were worried similar pages would cannibalize each other.
instead of blindly removing them, they analyzed thousands of pages + the exact queries each page ranked for.
they found the pages were different enough to target separate intent.
so they removed the blockers and switched the pages they wanted indexed to self-referencing canonicals.
that instantly opened a huge chunk of ColdIQ’s existing site to Google + Bing instead of creating more pages from scratch.
2. Bing was part of the strategy specifically because of ChatGPT
this is something a lot of people overlook
they made sure that their pages were indexable in Bing too because AI systems can pull fresh information through search infrastructure.
so the technical work wasn’t just “SEO cleanup.”
it directly affected whether ChatGPT could even discover the pages they wanted recommended.
3. Their content research pulled from places most SEO teams don’t combine
they used:
-> Ahrefs + Semrush + DataForSEO
-> SimilarWeb audience overlap + channel mix
-> Google Trends + Glimpse + Exploding Topics
-> live citations from ChatGPT, Perplexity, Gemini + Google AI Mode
-> Reddit + Quora + G2 + Capterra + TrustRadius
-> sales call transcripts
-> support tickets
-> live-chat logs
then they ranked opportunities by expected pipeline impact.
not just search volume.
that meant a question with lower traffic but much stronger buying intent could beat a huge keyword in the roadmap.
4. They turned ColdIQ’s own experts into the content channels
regular interviews were done with ColdIQ’s operators.
then those interviews were combined with:
-> LinkedIn posts
-> internal docs
-> existing content
instead of asking AI to invent “expert” content from scratch, they fed it actual internal knowledge buyers were already responding to.
the AI helped structure + scale it.
the expertise came from the company.
5. They reverse engineered the exact third-party sites shaping AI answers
for the prompts they cared about, they scraped:
-> which domains got cited
-> which sources kept influencing the answer
-> which third-party sites affected brand mentions + sentiment
those domains then became outreach targets.
so instead of only asking:
“what should we publish on ColdIQ (dot) com?”
they also asked:
“which websites does ChatGPT already trust for this question, and how do we get ColdIQ represented there?”
6. Reddit wasn’t treated like a traffic channel
they looked at which Reddit threads + subreddits were already getting cited by AI.
then they engaged specifically in those conversations and even started new discussions around attributes they wanted associated with ColdIQ.
the goal wasn’t Reddit clicks.
it was influencing the source material AI systems were already using to understand the category.
7. Their attribution system separated what they could PROVE from what they could only infer
AI search attribution is messy.
someone can discover ColdIQ in ChatGPT, close the tab, then Google “ColdIQ” 2 days later.
analytics calls that Google/direct traffic.
so they tracked:
-> first click
-> last click
-> assisted conversions
-> what prospects said on sales calls
-> what they reported during onboarding
-> CRM attribution
then they kept 2 numbers:
“clean attribution floor”
vs
“realistic influenced attribution”
clean attribution alone showed:
-> 15,700 visitors
-> 362 conversions
but when accounting for the direct lift, the estimated influence was much larger:
-> 55,813 visitors
-> 1,431 conversions
this is a much more honest way to measure AI search than pretending every customer clicks straight from ChatGPT into checkout.
8. The results
-> $506K in signed contract value
-> LLM search traffic grew 7.6X
-> organic search traffic grew 7.3X
-> SEO + AI search became their #1 traffic source
-> traffic share doubled from 16.5% to 33.3%
-> ranking keywords increased 983%
-> direct traffic increased 93%
-> ColdIQ went from 0 AI visibility to being recommended across dozens of high-intent prompts
the biggest takeaway:
don't just “publish AEO content.”
they unlocked pages that were already buried, fed real buyer language into the system, tracked the external sources AI already trusted, and measured AI discovery beyond last-click attribution to just name a few of the things they did.
if you want us to find where your company ranks + turn AI search into pipeline,
reply “BRR”
2
48
6,844
hoollly sauce
Webflow, valued at $4B, increased AI-referred signups to its template pages by 485% quarter over quarter.
here’s the AEO workflow behind it:
They made existing pages more useful
-> They identified 1,500 template pages to optimize
-> Added copy near the top explaining who each template serves and what it helps them do
They answered more specific questions
-> Went beyond “best CMS”
-> Created answers tailored to particular teams and integrations
-> After three weeks, share of voice improved for 70% of the questions tested
They made answers easier to find
-> Added tables of contents
-> Rewrote headings as questions
-> Tested these changes against control pages
They joined relevant Reddit conversations
-> Monitored threads connected to priority queries
-> Contributed helpful, transparent responses
-> Tracked the reaction
the reported results:
-> Template-page AI signups grew from 168 to 983 quarter over quarter
-> Overall LLM share of voice grew from 33% to 64%
-> The share of signups attributed to LLMs doubled from 4% to 8%
my takeaway:
before writing another article, inspect the pages closest to a purchase.
can a buyer quickly understand:
-> who this is for?
-> what problem it solves?
-> whether it fits their specific situation?
those are useful questions for your next AEO experiment.
pick a group of pages. improve the answers. track what changes in AI visibility, referral traffic and signups.
want us to find those opportunities on your website?
we’ll show you where you’re missing from AI search and what we’d prioritize fixing:
reply “audit”
crazy sauce
steal our favorite AEO hack:
-Ask ChatGPT a buying question your ideal customer would ask
-Think “best [category] for [use case]” or “[competitor] alternatives”
-Open the sources it cites
-Find the ones mentioning your competitors but missing you
this gives you a list of pages AI is already using to answer questions about what to buy.
basically a distribution map hiding inside the answer.
start with the sources that show up across several related questions.
then give those publishers a concrete reason to include you: a relevant integration, a missing use case, transparent pricing, or original data.
if it’s your own page being cited, make sure it clearly explains who your product is for and how it compares.
you’re finding the sources already shaping the shortlist and working to earn a place in them.
track whether your brand starts appearing across those same questions over time.
7
1,378
literally just tried it the other day it found so many security issues
thank god I was able to fix them before bad things happened
Introducing Compound Security
An open-source skill that helps your coding agent find vulnerabilities, patch them, and test that your app still works.
Here’s how you use it:
→ Help: figure out where to start
→ Brainstorm: pick what matters
→ Patch: plan, fix, and verify
→ LFG: take it through the workflow
and more :)
ppl in sf be talking about 996. i can guarantee you no one actually works 996.
its more like
9-10 doom scroll in bed
10-11 freshen up + breakfast + more doom scrolling
11-3 actual work (scrolling twitter)
3-5 work for 10 min, doom scroll, repeat
5-9 remember u forgot to eat lunch + talk to ur ai girlfriend
and repeat
i've noticed in the tech sf founder circle ppl mostly build for social status within the bubble.
meaning even if its not something they are actually interested in, or have some unique insight on, they will persue it just cuz it sounds cool to their founder friends.
everyone has a unique advantage in their life that they can use to find some niche problem to solve instead of hopping on the next trend.
so much sauce
Juicebox has generated 4,500+ new signups from AI search in JUST 2 months.
this is the insane AEO system behind it:
1. They reverse engineered the types of pages AI kept trusting
their research found AI systems were heavily favoring:
-> listicles
-> comparison articles
-> pages with really tight semantic alignment to the exact question
so they stopped treating every content format equally.
if AI kept pulling from comparison pages for a certain type of search, they built comparison pages.
if listicles kept winning, they built around that format.
2. That research turned into a very specific content plan
instead of just “publish more blogs,” the plan included:
-> competitor comparison pages
-> listicle-style articles
-> homepage content changes
-> schema improvements
-> third-party citation work
-> a new /ai-recruiting page
-> a new /peoplegpt page
every page had a specific gap it was supposed to fill.
3 They only needed a handful of pages to start dominating dozens of searches
they created 20 new articles, guides, and competitor comparisons.
5+ of those blog posts eventually became top-cited pages on Perplexity across 40+ high-intent recruiting searches.
the lesson here is pretty insane:
you don’t necessarily need hundreds of pages.
a small number of pages that perfectly match what AI wants to retrieve can win across a huge number of searches.
4. They also went after the searches closest to a buying decision
not just broad searches like:
“best AI recruiting software”
they specifically targeted:
-> “Juicebox vs [competitor]”
-> “[competitor] alternatives”
-> specific recruiting problems
-> specific hiring scenarios
so Juicebox could show up when someone was already comparing products and much closer to buying.
5. Third-party websites were treated like part of Juicebox’s own AEO strategy
they didn’t only optimize Juicebox.com.
if AI kept citing certain external sources for an important search, those sources became part of the strategy too.
the goal was to increase the amount of information across the web giving AI a reason to recommend Juicebox.
not just the information sitting on their own website.
6. They even optimized the way AI talked about the brand
they tracked:
-> negative talking points
-> advantages AI kept giving competitors
-> recurring narratives around Juicebox
then worked on correcting those narratives across the information AI was using.
their sentiment score went from:
2.35 -> 3.1 out of 5
so AEO wasn’t just about getting mentioned.
they were trying to control HOW the brand was being described.
7. They treated model updates like Google algorithm updates
when ChatGPT moved from GPT-4o to GPT-5, the strategy was adjusted within 2 days.
because whatever gets cited by one model today isn’t guaranteed to keep winning after the model changes.
that means the playbook itself can’t stay static.
8. The results
-> 4,500+ new signups directly from AI search
-> ChatGPT traffic increased 336%
-> Perplexity traffic increased 234%
-> 20%+ of inbound traffic started coming from LLMs
-> Juicebox became the #1 cited domain across its priority recruiting software searches
-> Juicebox had 161 citations, ahead of YouTube at 119 and every direct competitor
-> in AI sourcing + people search, Juicebox appeared in 85.4% of relevant AI searches
the biggest takeaway here is:
they figured out which page formats AI already trusted, built specifically around them, targeted searches closest to buying, expanded that beyond their own website, and even optimized the narrative AI had about their brand.
if you want us to find where your company ranks + get you customers from AI search
reply “audit”
3
7
2,855
hooolllyyy sauce
Ramp, a $60B company, increased its AI search visibility in accounts payable nearly 7x in JUST 1 month.
this is the genius AEO system behind it:
They identified a specific category where they were barely showing up
-> Ramp’s accounts payable AI visibility was just 3.2%
-> That gave them a clear category to focus on and a baseline to measure against
They studied the content AI was already citing
-> Automation, AI and software comparisons kept appearing
-> Only 6% of Ramp’s existing citations came from automation-related content
-> This revealed an opportunity to expand their coverage
They created pages for specific buying decisions
-> Accounts payable software for small businesses
-> Accounts payable software for large businesses
-> Two different buyers, with different requirements, getting dedicated answers
They supported those pages with related content
-> Accounts payable automation software comparisons
-> AI in accounts payable
-> Together, these pages covered more of the questions surrounding the purchase
the reported results:
-> The two targeted pages generated 300+ citations within one month
-> AI visibility increased from 3.2% to 22.2%
-> Ramp climbed from 19th to 8th among fintech brands in the accounts payable category
the lesson for your company:
your next AEO opportunity could be a buying question your website hasn’t answered properly yet.
start by asking:
-> Which questions should our product show up for?
-> Who gets recommended today?
-> Which pages are being cited?
-> Do we have a page that actually answers the full question?
then turn those gaps into specific experiments:
-> choose a buyer question
-> improve the relevant page or create one where coverage is missing
-> rerun the same prompts
-> measure citations and mentions
-> track whether those gains bring qualified visitors and customers
visibility gives you a signal. customers tell you whether it’s working.
want to know where your company is missing from AI search?
if you want to see where ur ranked
reply “audit”
sf is the only place where ppl will unironically dress up as a claude code limit reached message for halloween
5
20
835
so much sauce
Descript, a $550M company, increased subscriptions 56% in JUST 30 days using AI search.
this is the genius AEO system behind it:
1. They started off with 1,000+ pages they had already been published
-> 600+ blog posts
-> 400+ tool pages
instead of pumping out another 100 AI-written articles, they focused on making the content they already owned perform better.
and there was a lot to fix.
their old programmatic content system could produce pages fast, but quality was inconsistent and older pages were starting to decay.
2. Before rewriting anything, they cleaned up the content library
they checked for:
-> pages competing against each other
-> duplicate content
-> outdated pages
-> weak titles + descriptions
this is important because publishing a “better” article means nothing if 3 other pages on your site are fighting it for the same topic.
they fixed the structure of the library first.
3 Then every page went through a keep / improve / remove process
instead of telling AI:
“rewrite this article”
their workflow first figured out:
-> what information was still useful
-> what was outdated
-> what was missing
-> what should be removed entirely
so their AI wasn’t generating a new article from scratch every time.
it was making targeted changes to an existing page that already had authority.
4. Then they changed the pages specifically for AI search
this was one of the biggest parts.
Descript added authoritative sources to support claims.
they restructured headings so the page was easier for search engines and AI systems to understand.
and they updated old statistics + research instead of leaving stale numbers sitting inside articles for years.
AirOps measured Descript’s average AEO score before and after this.
before: 61
30 days later: 78
a 28% improvement.
5. The outbound sources weren’t random links either
AirOps’ research found pages with multiple authoritative external citations were getting cited by AI systems far more often.
so instead of making unsupported claims like:
“video is becoming the most important content format”
the content would point to the actual study, dataset or expert behind the claim.
that gives an AI answer something it can verify before repeating it.
6. They also made the content easier to rip answers out of
long walls of text got broken into:
-> clear H2/H3 sections
-> short answers
-> lists
-> direct explanations
-> question-based sections where relevant
the goal was simple:
if ChatGPT needed one paragraph from the page to answer a question, that paragraph should make sense without needing the other 2,000 words around it.
7 Then they solved one of the biggest problems with doing this at scale:
brand quality.
Descript baked its own brand guidelines into the workflow.
so hundreds of pages could be refreshed without every article suddenly sounding like generic AI content.
the workflow handled the repetitive research + rewriting.
Descript still controlled what quality or you can sya they controlled what “good” looked like.
8. They put QA inside the system instead of doing it manually at the end
the workflow could check things like:
-> brand consistency
-> outdated information
-> content gaps
-> metadata
-> structure
-> duplication
before the refreshed page went live.
that helped them refresh content 9X faster without adding a giant content team.
9. And they didn’t treat a refresh as a one-time project
this might be the most useful part to steal.
the system could monitor performance and trigger another refresh when a page started declining.
so instead of:
publish -> forget about it for 2 years
they built:
publish -> monitor -> decline detected -> audit -> refresh -> measure again
content became something they maintained continuously.
10. Then they measured whether the refreshes actually changed anything
Descript’s average AEO score went:
61 -> 78 in 30 days
their organic traffic increased 35%.
their content refresh process became 9X faster.
and their brand visibility score reached 28% after the refreshes.
but the number that matters most for every brand:
subscriptions increased 56% in 30 days during the period Descript says corresponded with content engineered for AI agents.
the main takeway here is NOT to publish more AI content.
instead of constantly publishing new pages, Descript kept finding which of its existing 1,000+ pages were losing performance, fixing them, and then building/iterating that system.
if you want us to find what’s stopping your company from ranking & want to rank #1 on AI search,
reply “audit”
3
16
4,137
this changes things in so many ways
Introducing AgentWeb: reliable web infrastructure for generalized agents.
47/47 real world tasks on live sites. 1.9s median. $3 total.
Request access: agentweb.us
every city has a vibe check
sf: "what are u building"
nyc: "what do u do"
la: "who do u know"
austin: "where did u move from"
dubai: "what do u drive"
sf is still the only city where ur startup idea is how you introduce urself
2
11
1,300
so much sauce
Granola, a $1.5B company, increased monthly new users from AI search by 4,600% in JUST 7 months.
here’s the genius AEO workflow behind it:
1. They started with buyer decisions, not keywords
-> Granola mapped the questions people ask before choosing an AI meeting tool
-> Then they turned those same decisions into both Google keywords and AI prompts so they could measure where the brand actually showed up
2. Every lost prompt got diagnosed before anything was created
-> If Granola wasn’t showing up, they first asked why
-> Was the right page missing?
-> Did the page already exist but fail to answer the question properly?
-> Was there a crawl/indexing problem?
-> Or was AI getting the answer from third-party sources that barely mentioned Granola?
3 They separated “missing page” from “weak page”
-> This sounds obvious, but it changes the entire content strategy
-> If a relevant page already existed, they improved that page instead of publishing another article targeting the same question
-> New pages were only created when there was actually a gap
4. They stopped treating every technical SEO issue like it mattered equally
-> Crawlability, indexing and performance problems were prioritized based on whether they affected pages tied to real buyer demand
-> A technical issue on a page connected to an important buying question came before cleaning up random site-wide warnings
5. They studied the exact third-party sources AI was already trusting
-> For prompts Granola lost, they looked at the Reddit threads, review sites, articles, directories and other pages showing up instead
-> The question wasn’t just “how do we get more backlinks?”
-> It was “which sources are shaping this specific recommendation, and why are competitors represented there better than us?”
6. They used a simple framework for every important source:
-> OWN: pages Granola controls directly
-> CLAIM: profiles and listings they can take ownership of
-> EARN: editorial mentions, reviews and community discussions they have to deserve
-> CONTEST: outdated or incorrect information they need corrected
-> Different source = different action
7. Reddit became part of the search strategy
-> Instead of posting everywhere, they focused on communities much closer to the product and buyer
-> The same buyer questions being tracked in AI search were used to decide which conversations were actually worth joining
8 They measured Google and AI separately
-> A page could rank better on Google and still barely move in ChatGPT
-> AI mentions could rise while organic traffic stayed flat
-> So Granola tracked both systems independently instead of assuming SEO performance meant AEO performance
9. Then they reran the same questions and scaled what actually moved users
-> measure the gap
-> diagnose why it exists
-> make one specific change
-> rerun the prompts
-> track whether visibility and activated users moved
-> double down on what worked
the result:
-> monthly new active users from AI search increased 4,600%
-> non-branded AI citation rate increased 558%
-> non-branded mention rate increased 739%
-> organic sessions increased 134%
Granola didn’t win AI search by JUST publishing more blogs & content like every other company
instead they built a system for figuring out exactly what they were missing, choosing the right fix, and measuring whether it actually brought in users. then after that just iterating and doubling down on what worked
AI search is now one of their biggest acquisition channels.
and it should be one of yours too!
if you want us to run an audit for your company and make you rank #1 in AI search,
reply “audit”
