Pinned Tweet
Today, we’re introducing a new way to win in AI search.
If we don't make you more visible, don’t pay us.
No $5k/mo dashboards to stare at.
No recommendations for your team to implement.
We execute everything.
And we put our money where our mouth is:
a) Pay when an agreed visibility score is reached
b) Pay a percentage of the incremental business we generate
We take the risk. You get the upside.
One partner per market.
Spots now open!
people still underestimate how powerful specific product/service recommendations from LLMs are
LLMs are deemed a source of authority so them mentioning you leads to empirically higher conversion rates than any other acquisition channel
and the best way to win in this new, emerging segment is to have a consistent footprint across the web about your brand
providing a great service and having truly happy customers does wonders for that!
The products that LLMs recommend will be the dominant products people use.
You win this game by:
1. Building a product that customers love and talk about on the Internet.
2. Playing well with LLMs, e.g. with API/MCP/CLIs etc.
Great to see @Mercury wining here.
How does ChatGPT (and other LLMs) see your site?
Get visibility into the 'mind of agents' when they decide whether your brand is cited, recommended or skipped
if ChatGPT can't read your site, it just recommends someone it can, and you may never even find out it happened!
so we built a free tool that loads your page the way GPTBot and other LLMs do
and show you everything it trips over, from a file that blocks them, to pages that stay empty until a specific script runs
Paste a URL into the Agent Readiness tool for free and get a visualization of what those systems actually retrieve
algomizer.com/tools/agent-re…
JEV reduced our AI Search / GEO evaluation costs by 54%
what does it do?
1) Ingest answers from the engines buyers use
- we collect consumer-UI answers on a fixed prompt suite. Jev judges them
2) Named recommendation scoring
- each row becomes: named / not named, pick / mention / absent, confidence attached
3) Ghost-citation detection
- marks answers where we fed the model and never got the name (the failure mode most mention dashboards hide)
4) Incumbent displacement tracking
- tracks whether the shortlist still defaults to the same titans, prompt by prompt, run over run
5) Persona split labeling
-same question, different buyer framing- Jev buckets which persona got which brand list so we can see instability instead of averaging it away
6) Change-impact checks
-after we touch a source or entity signal, it re-scores only the affected prompt cluster and tells us what flipped
7) Human review gate
- drafts and edge cases pass a closed checklist, low-confidence or failed checks go to an operator, the rest stay in the log
Incredible efficiency and savings across the board
we just finished a 90 day GEO experiment that most people in AI search will hate
we took a partner from 4% to 80% visibility, pushed BlackRock out of the answers (with decades of authority and $15T under management)
and didn't publish a SINGE piece of CONTENT the entire time
no blogs, no landing pages, no "AI optimized" articles, literally nothing
now our partner is showing up ahead of multi-billion dollar legacy brands like BlackRock, KKR and Temasek in LLMs
I want to explain why we ran it this way, because I think most people in GEO are spending their money on the wrong thing
if you look at every tool in this space, the product is almost always *content*
if you look at every agency pitch, it's some version of "we'll write 40 articles a month for you"
and the reason is quite simple
content is simply the easiest thing to produce, sell and the easiest thing to invoice for
but what nobody had actually tested and questioned is whether it's the thing that moves the needle
or whether it just happens to be the thing everyone is doing!
so we picked a brand new partner, and didn’t deploy a SINGLE piece of content for 90 days
and pulled on every other lever we had
(we really wanted a clean answer)
now, here's the thing you need to understand about how these models work (and think)
instead of reading your blog and deciding you're the best in your category
the LLM builds a picture of you from everything it can find
and most of the weight goes to what other sources say about you, not what you say about yourself
your own content is literally the least trusted input, because you're the one claiming you're great and the model knows that!
so instead of writing, we spent the 90 days on the stuff that actually feeds that picture
1) the first thing we fixed was the entity itself
the partner was described in six different ways across the web depending on where you looked, and models punish that kind of inconsistency hard.
we went through and made every source agree on who they are, what they do and who they serve.
2) then we mapped the sources
for every buyer prompt in the category, we pulled which domains the models were actually citing, and then made sure the partner was present and correctly described on all of them.
directories, databases, industry lists, wikis, news.
if a model pulls from a source and you're not on it, you essentially don't exist for that question
3) after that we worked on third party validation, so mentions and references on sites the models already trust, framed around the actual questions people ask
we also cleaned up the structured side, schema, naming, entity data anywhere a machine reads it.
boring but it matters!
4) and the whole time we tracked the real buyer prompts, reverse engineered the sources behind each answer, and went after it one by one
the combination of all these is what took them from 4% to 80%.
companies with billions in brand spend were simply pushed out of the answers by someone who didn't add a single new page to their site!
think about that next time someone's convincing you that it's all about content
also, the lesson isn't that content is useless
content works, but only once the model already believes you're credible!
most people are doing it backwards and building the roof before the foundation exists, and then wondering why 40 articles a month isn't doing anything meaningful
if you want the exact playbook we used on this, RT + comment "Visibility" and I'll send it over
The absurd amounts of money I would pay for something that solves attribution for traffic derived from LLMs (ChatGPT, Claude, Perplexity) etc.
the final form of all AEO / GEO is content marketing
why?
AI has made writing as a format extremely abundant, cheap and available
this leads to the extreme slopification of anything and everything around us (e.g. thousands of blog posts can now be made in seconds for free)
so for the LLMs to remain 'valuable' and filter through noise, they are essentially forced to lean into either:
a) authority
b) UGC platforms with stringent moderation & controls
c) a mix of the above
and this is exactly what Reddit had and why it ended up being deemed a highly authoritative source in ChatGPT
but because it did, it eventually started becoming abused for those exact purposes
this is a cat-and-mouse game and a flywheel that has always been prevalent in the world of search optimization and that is unlikely to go away
so there will always be 'low-hanging fruit' in short term, opportunistic plays that you can lean into, but that you should never ground your entire strategy in
that being said, social media platforms like YouTube, Reddit, X, LinkedIn, Facebook will increasingly become a source of reference for LLMs, as they tick all the right boxes
they
a) have their own spam-filtering mechanisms
b) get continuously better at demoting AI slop
c) increasingly lean into rewarding real, identifiable, valuable, human content
d) have developed thorough algorithmic weights over decades
instead of building those out from scratch and spending billions of dollars + decades learning
the frontier labs can simply tap into all of that by just nudging the LLMs to deem the content from those sources more authoritative and reputable
there are very few incentives for them to not do so
we are starting to see it with the rise of YouTube in LLMs now
so get your social media marketing strategy in place, ensure it is comprehensive, consistent and value-additive
and soon you will start reaping the 'indirect, second-order effect' rewards from those initiatives
that might eventually outweigh everything else
the earlier you realize that and the earlier you act on that, the better your AEO / GEO strategy will be
A lot of discussion on AEO measurement recently.
I haven’t read too much I think is wrong but it is missing some key things from my pov.
In GTM the only thing that matters is what did you spend and what did you get. Revenue is the KPI.
But the thing is most of how you drive ai visibility is through the secondary effects of other types of marketing you are doing.
A single page might do work across seo, sem, aeo at the same time. YouTube. Reddit. G2. Each investment now has direct outcomes and indirect outcomes (aeo) but that cost is just one line item.
This is one of the central reasons why I believe that almost everything you do in aeo you would do for other reasons. Because investments that have multiple dimensions of value typically return better and have more durable value.
My biggest concern with most of what I see in ai visibility today is that if you say the only queries that matter are consideration set building queries. And ~~right now~~ chatgpt fans out on those in a way where it asks the question then stuffs 5 brands and 2026 in there for good measure, the only way to intercept that query is with a listicle where you put yourself first.
The result is that too much of the marketing for this channel has no other value. And we are so early that it will surprise no one if reddit vanishes (kinda happened already), if fanout structure changes, etc. Maybe everyone is ok with chasing whatever hack still works but if thats part of your strategy it should probably be a minority of it.
I would do a few simple things instead.
1) put one team in charge of your core owned media engine (or at minimum seo, aeo, web) and the official financial thing I’d track is the cost and return of organic/direct/ai referral. I might discount it a little because branded organic/direct obviously has more than that. But I’d mostly do that. Invest and measure like you know one thing can do three things.
2) broad based ai visibility indicators eg things like ahrefs brand radar, ai referral traffic and bot traffic - ai referral traffic and bot traffic is real data not made up data… they have noise too but if youre trying to understand what parts of your site are actually visible and relevant, this data is how to answer the question… ahrefs brand radar gives you a data set where individual prompt data is sparse but the library is vast, closest thing to traditional site reporting in seo and unless you are insanely diligent about your prompt tracking is likely much closer to giving someone feel for their overall visibility
3) track the full funnel in ai prompt trackers… citations at top, mentions in middle, win/recommendation rate and some view of sentiment by category heuristics… i actually am not a huge fan of audience derivatives because when i set up tracking i prefer to derive prompts from keywords so i can pair them with volume assumptions and many role/use case scenarios make it into keywords… so its not so much abt prompts derivatives as just different keywords
Are some of these measures more/less important than others? Sure obviously.
But if I’m brand the thing I want is pervasive visibility. I want to be extremely visible full funnel in organic search. I want full funnel visibility in ai engines. I want real observable metrics like ai referral traffic and bot traffic going up (as indicators). I want the aggregate cost and return to make sense. I want to be doing really good full funnel marketing across all agent legible surfaces. And I want to measure all of it.
If all that is happening I am good.
If none of that is happening but I rank on commercial prompts in chatgpt because of some load-bearing listicles? Id feel not good.
Anyway like I said nothing I’ve read I think is wrong, just not exactly how I think about it.
Use Grok for shopping and it will literally save you money and time.
It is insane in shopping and finding the best deals.
Shopping is arguably the hardest task for an LLM
for a prompt like "find me the best deal on XYZ", the AI needs to:
1) choose what to search
2) judge which sites to settle the question through
3) check the prices that move daily
4) know when there's enough to answer
no single page holds the answer, so the model has to build one out of many sites!
and Grok just proved how capable it is in saving you money
After pulling the retrieval logs behind many shopping questions to see how it actually builds a purchase answer.
For one deal question it had:
- Run 41 separate searches and page-opens
- Read 18 full pages instead of skimming the previews
- Checked a year of price history to see whether today's price was actually low
- Looked at resale, refurbished, and the forums where people post discount codes within hours of finding them
- Guessed four prices it expected the headphones to be selling at, then searched for those exact numbers to confirm somebody was really charging them
- Pulled Amazon's live prices off the page as they were at that second
Then it threw almost all of it away.
142 URLs retrieved, yet only were 5 cited in the end.
No special instructions. No plugins. No shopping API.
If you change the wording, it'll change who it trusts.
Ask which is best and it goes to the labs that test them.
Ask where to buy and it goes to the shops.
Ask for a deal and it goes to the price trackers.
It knows which kind of source can settle which kind of question, the way an experienced buyer does.
Grok 4.5 opens twice as many pages as Grok 4, finds sellers the older model never reaches, and still names fewer sources, because it reads more so it can throw more away.
one underappreciated aspect of AI, is that its *supercharging* agencies
by effectively turning them into a productized service providers
and enabling them to charge for outcomes
the greatest impediment for that model was always scaling and pricing
but both are ultimately redefined now
the more verbose the industry, the greater the usage
AI power users are showing up outside tech. The fastest-growing Codex adopters since February:
Legal: 108x
Sales: 41x
Recruiting: 41x
Marketing: 26x
Healthcare: 24x
Charts of the Week: a16z.news/p/charts-of-the-we…
are agents using far more tokens because it is in the best, genuine intetest of finding the answer/solution?
or because they are inherently incentivized to do so?
we'll find out with time
at this point it's worth inverting and asking
is there a single AI data-labelling platform NOT growing at this pace?
The team at @micro1_ai is on fire.
$500 million run rate now.
Up from $100 million just 8 months ago.
@aliansarinik lets go!
the first mover advantages of being entrenched in these platforms that are *clear winners* already are massive
and almost always leads to generational, multi billion dollar business emerging
it was the case with early builders on App Store, Wordpress, Shopify and many others
just by tapping into the tailwinds, growth and velocity of the underlying behemoth
a rising tide lifts all boats!
would be awesome to see how much of this revenue can be attributed to the ad with Rick Hoffman
7/10 people I talk to now know about Pocket just through that alone!
This is why we always emphasize not approaching AI search visibility as a "one and done" effort
something definitely might be *more performant* than everything else, at a given point in time
and its completely fine to lean in!
but make sure you have a healthy foundation in place all around
so that when these changes do come about, you are not wiped from one day to another
looks like reddit is almost wiped from chatgpt sources
the query fanout changes had a big impact
and the past couple of days it seems to be almost completely removed from prompt responses
promptwatch.com/data/reddit-…
Margatix retweeted
Today, we’re introducing a new way to win in AI search.
If we don't make you more visible, don’t pay us.
No $5k/mo dashboards to stare at.
No recommendations for your team to implement.
We execute everything.
And we put our money where our mouth is:
a) Pay when an agreed visibility score is reached
b) Pay a percentage of the incremental business we generate
We take the risk. You get the upside.
One partner per market.
Spots now open!
AI search is empirically the highest-intent acquisition channel on the market right now.
Every day, more than a billion people ask ChatGPT, Claude, Gemini and Perplexity which product to buy, company to hire, or solution to use.
We put our partners in those answers.
Become an @algomizercom partner today:
algomizer.com
