Fall is about to turn everyone’s brains back on.
I’ve enjoyed the long sunny days and the pace of summer. But when kids go back to school, and pumpkin spice lattes are on the menu…you know what that means: it’s time to get into the zone!
This has been a growth summer for me as I developed Surfaced, using AI systems to develop my ‘context layer’, pushing things to the limits to see how to operate my own business and leverage my experiences and brain to help develop frameworks to help businesses.
Using that as an education, I feel a deeper understanding of how to work WITH AI for search. Got to be in it to help with it.
This summer we were bombarded with data, opinions, technology changes, blackouts, heroes and villains…we’ve never had more to consider, and it’s not slowing down.
Onward!
Google put a turnstile on its own search results. Wow.
Just last week, Google confirmed it is rolling out google.com/goto passthrough URLs in search results. The link behind a result now points at a Google redirect instead of the destination page. Google's only statement was about deploying technical measures against evolving forms of abuse. Ya ok.
It did not explicitly call out name scrapers, rank trackers, or AI companies - but those are all feeling the burn.
Goto links cannot be decoded (people have tried), so tools have to follow every redirect. Research shared that it takes 500 to 1,000 requests to resolve a five-page ranking, since each SERP carries hundreds of links. That’s a lot of money to spend that will cost tool providers.
Note: DemandSphere noted last week as well that signed-in users may or may not get the “goto treatment”, and pointed to incognito as the reliable way to reproduce it.
For marketers? Nothing to do on your site, but be prepared to look at your measurement tools going potentially wacky.
My take: Google is trying to protect their entire web crawling system, which they’ve put hundreds of millions of dollars into developing. Remoras have been leeching off them for a long time. As partnerships deals are made and everyone is trying to position themselves stronger in the future, we’re seeing things like this and Reddit disallowing crawling in their Robots.txt as warning signs.
What this means: Rank and crawl data will get more expensive because of this, so monitor your SaaS prices…some vendors will eat the cost until they can’t. Just Yesterday, Tim Souro of AHREFs just posted a “we’re figuring this out, thank you for your patience” on LinkedIn.
Reddit fell off a cliff in ChatGPT, and the whole internet decided to figure out why
Per Promptwatch, Reddit's share of ChatGPT Search citations averaged 3.83% from July 18 to August 7, then collapsed to 0.52% between August 14 and 17. That is an 86.4% relative drop…which is a lot.
For context, as recently as April, Reddit was the single most cited domain in ChatGPT Search at ~4% of all citations.
Google AI Overviews declined roughly 11% over the same window and AI Mode roughly 30%.
There’s a lot of good research into why Reddit is still very important and used in RAG and informs answers, the bottom line is that currently they are not being cited as much.
It’s still very much heavily influencing answers, they are just not getting credit in your tools. Don’t jump ship.
What to do? Diversify, diversify, diversify. A third party can end your citation share overnight, with no ranking signal, no warning, and no appeal, as we’re seeing. Make sure your action items are not fully invested in one place.
If your visibility depends on one singular platform (like Reddit) you don't own, you are merely renting on borrowed time.
Most of the socials in the last 2 weeks were around Reddit dropping and Google’s go-to pass-throughs, tons of opinions floating around.
IAB released a framework because 20-plus vendors were selling AI visibility scores with no shared definition of a mention or a citation, and they were getting tired of it.
I know this all too well, having been on the inside here working at multiple platforms.
They found 16% of brands track AI search visibility in any systematic way (McKinsey CMO surveys), which means every pretty tool you pick has a different way of measuring.
What this means: you need to understand how your tool is calculating its scores because there is no consensus across the lot of them. Be defensible, and if you change providers, understand the change.
The Reddit crash got a Forbes treatment on August 20, which means your C-suite saw it…maybe you even got some panicked emails.
I attended Similarweb’s Future of Search with Lil Ray, Kevin Indig, and Ethan Smith (what a lineup) and wrote about it on LinkedIn, but the tldr here:
The core model/training data is more important at creating short lists than live web retrievals, meaning it will take time for your efforts to have its true impact.
AI Mode is coming: evolution of the results and the adoption of users as they become more familiar, and the data shows it’s happening whether we like it or not.
Citations matter, but how you are mentioned is more important than the attributable link (data says nobody clicks them). Who cares if you’re referenced when the answer is bad.
A thing happened and people can’t agree on it.
Promptwatch measured Reddit's ChatGPT citation share falling 86.4%.
A separate Searchable analysis put it at 83% between August 9 and 18, which lines up. Close.
But over that same window, Searchable found Reddit usage increasing 26% in AI Overviews, 32% in AI Mode, 6% in Perplexity, and 24% in Gemini.
How the heck does that happen?? Different platforms cite differently, so that’s one answer.
Another one is more practical though, the thing I keep shouting: your visibility number is an artifact of your prompt set, not a fact about the world. If two vendors can look at the same three weeks and hand you opposite directional reads, then the number is just a proxy and not a fact.
Organizational maturity is the thing very few are talking about
Tom Critchlow and Mike Mallazzo published an AI Search Maturity Matrix on August 21. They opened therir article with a company spending $500k a year on prompt tracking, thousands of prompts, and a weekly all-exec email containing a citation rate line chart that wiggles.
Honestly, this is the whole industry in one line.
It is the best public attempt yet at treating AI search as an organizational capability instead of a dashboard that floats around inboxes. I loved that they shared this to bring more awareness, and it deserves it.
I have something similar but am keeping it for my advisory practice 🫠
They took a real hotel, ran hundreds of simulated AI recommendation queries against four real competitors, and changed only the wording of the listing each time. Reordering the copy to lead with what business travelers care about moved the model's recommendation rate from 16% to 78%(!).
But removing the brand names barely moved it. The words drove the choice.
A finding is only as good as the org‘s ability to act on it.
As a consultant, I can’t tell you how many findings from measurement platforms are mistaken for thinking you have a strategy.
It’s not that measurement is garbage, but it’s that everyone thinks it’s a saving grace. Profound is leaning into engineering, Scrunch into agents, Bluefish into accuracy, Evertune into data. They all have their place, and it’s not shade - it’s just not what companies actually need to go up and to the right.
Audit what you're renting
So if your visibility depends on a platform you don't own, you're renting.
This doesn’t take much time, and it's a useful exercise you can do in the next two weeks, because Cloudflare's default bot blocking change lands September 15!!
Pull your top 20 cited sources from whatever tool you're using. Sort them into three columns: owned (your site, your docs, your app), influenced (review sites, directories, partner pages, anywhere you have a profile and a login), and rented (forums, editorial, aggregators, anywhere you have no account and no relationship).
Look at the rented column and ask what happens if it goes to zero. (see Reddit 😅 )
Check your own bot settings before September 15. Cloudflare flips defaults for new and free tier customers that day. Then check whether the influenced sources that cite you are behind Cloudflare too, because if a directory you rely on starts blocking crawlers, you lose that citation path.
Email your rank tracking and AI visibility vendors this week asking the question: what does the goto change do to your collection method? Tim Soulo of AHREFs just addressed their data concern about this yesterday on LinkedIn.
Export a baseline including things like rank data, citation share, source mix, whatever you have. The collection methods underneath all of it are changing right now.
Measurement is extremely difficult in 2026.
"Where are we on the ai search maturity curve?"
I get some version of this when I mention it on calls, maybe not as explicitly, but Tom and Mike's work validates what I’ve been doing. Everyone wants to size up and see where they stand.
Their model has three dimensions: measurement capability, executive knowledge, and strategic posture, across five levels.
They define AI search maturity as the organizational ability to turn noisy model outputs into foundational insights about the information ecosystem, and then into durable market influence.
They are explicit that most companies sit at level two aspiring to level three, and that almost nobody is genuinely operational.
I’ve seen about the same in my journeys. I've been running my own maturity model for a while, built from a different angle, and the two disagree slightly:
1. Their model measures the program. Mine measures the organization that has to run it.
Measurement capability, executive knowledge, and strategic posture describe the AI search function. But the AI search function doesn't produce the answer - seven or more other functions do.
2. Level three is where everyone stalls, and it is not a search problem.
Tom says very few companies are truly operational, and I agree. Most fit in the middle.
Level three requires brand, content, product data, PR, community, and web to move on a shared finding, and not one of them reports to whoever is accountable for AI visibility. Most are still figuring this out.
3. A tier is aspirational until you ask how much of it is actually AI-augmented.
This is the layer I've added to my own operation, and it changes almost every time I run it(and honestly it’s a bear to keep up with the modeling changes that impact workflow). Most people are still explorers, me included.
4. Everyone overestimates by about a tier and a half.
Ask a search director about their AI search strategy, and you get a fluent and confident answer. Ask whats-his-name over there in product about what you’re focused on, and you likely won’t get an answer.
Maturity is not how smart the team or leader is, it is what the organization can say about you and what you’re doing when you’re not around.
I have been the problem myself inside enterprise organizations, and seen it from the outside being at platforms like Brightedge and Brandlight, looking at customers’ operations.
Obligatory CTA: if you want to understand where the maturity of your organic search program is, reach out. I am happy to help and give you my time.
Give me the Fall, I am done with this summer heat in Boston.
Looking forward to the sprint towards the holidays.
Next issue, I’ll likely be sharing some of the agents I’ve been creating to help with solving common challenges.
Until next time,
Andrew