My selection of the best AI Search tactics, tools, strategies, and stats to help you design your content strategy in the age of GEO. Like every month:
🟣 Top searches & data – AI Search stats to help you design your strategy
🟣 The AI Search tech corner – A “ugh” term from the tech side of AI Search
🟣 Search Everywhere Optimization – Beyond Google
🟣 GEO Strategies – GEO strategies for your Executive Committee
🟣 GEO Tactics – Field tactics from top experts
🟣 Reading of the month – The content that everyone is talking about
🟣 SEO vs. GEO – The naming war continues
🟣 AI Search Tools – The tools that matter ( mostly free )
Go! 👇 .
This month, two major benchmarks converge — Ahrefs’ first major AI Search report (100M+ data points) and the study quarterly Datos × SparkToro Commentary by Rand Fishkin.
Two studies, but the same verdict: AI grows quickly but remains small, Google dominates, and — against all odds — zero-click backs away.
AI doubles but stays tiny: less than 2% of desktop visits (0.41% to 0.93% in the US), and Google still sends 190 times more traffic than ChatGPT. In contrast, Google holds 94-95% of searches.
Counterintuitive: the zero- click declines (22.4% in the US, the lowest point of the period) and clicks organic rise (44.9%).
The real visibility signal is YouTube: out of 75,000 brands analyzed by Ahrefs, it is the source that correlates most strongly with presence in AI.
The AI race is getting tighter: ChatGPT is the leader (~35% of US AI usage) but stagnating, Gemini is number 2, and Claude is jumping to 8.5%.
And above all, search is fleeing search engines: YouTube, Reddit, Amazon, and TikTok are capturing it. a growing proportion of journeys.
These two studies will reassure you—wrongly so, if you only look at the big numbers. Yes, AI accounts for less than 2%, yes, Google sends 190 times more traffic. But volume isn’t the point: these figures measure the traffic sent, not the ongoing shift. The real signals are buried in the details—it’s YouTube that’s driving your AI visibility, and search is abandoning engines for Reddit, Amazon, and TikTok. Don’t read “AI is still small” as “I can wait.” Read it as “I still have a bit of a head start—for how long?”
Kevin Indig relays the largest behavioral study to date on Google’s AI features: 846,000 real sessions (clickstream data from Surfer SEO, analysis by Eric Van Buskirk), February-March 2026. His central finding: with an AI Overview on the page, we no longer read in a linear fashion.
Nearly half of all scrolling is now upwards (47.5% compared to 27% without AI). Users no longer scan, they reread.
A direct consequence of this is that your result no longer has a single impression, but two or three. The first impression triggers the reading (on scrolling down), the second triggers the click (on scrolling up).
The same user actually does the opposite depending on the interface: in AI Mode, they accept the answer and stop (88% use the shortlist as is, 64% don’t click on anything); in AI Overviews, they compare.
Brand search has lost its shortcut: even someone typing your name first scans the surrounding information before clicking.
Remember one word: “opposites.” “Being visible in AI” is starting to mean a lot of different things—in any case, behind “I’m visible in AI” there are at least two different jobs. Appearing on an AI Fashion shortlist is all about the model; winning the AI Overview comparison is all about the page itself, facing your competitors, under the watchful eye of someone who proofreads twice and therefore, in effect, compares. And the comfortable idea that your brand recognition protects you—that the most “well-known” brand wins—is no longer as true as it was just a few months ago: even someone who types your name in will first look at the competition. The brand opens the door. It no longer lets anyone in.
🔗 Users behave differently in AI Overviews vs. AI Fashion
Aleyda Solís analyzed nine months of data from the Semrush AI Visibility Index to compare where ChatGPT and Google AI Mode source their information across five sectors. The two platforms rely on profoundly different—and constantly evolving—ecosystems.
ChatGPT is 5.4 times more diverse: 1,621 brands reach 80% of the share of voice, compared to 299 for Google AI Mode.
YouTube became the number one source for Google AI Mode in April.
Reddit is the only truly consistent cross-platform brand; Forbes also remains very stable.
Everything is moving fast: Wikipedia (number 1-2 in ChatGPT since August) has fallen to 5th place; Alibaba has surged to 4th place (and number 1 in consumer electronics).
For good, stop using the singular form of “AI.” ChatGPT and Google AI Mode don’t read the same web: one scans everywhere, the other focuses on a handful of sources—and their menus change from month to month. The universal “AI visibility strategy” is a marketing myth. What you need is a map of the third-party sources that matter in YOUR industry, platform by platform—and the discipline to keep it updated, because what cited you in March might have forgotten you by April.
🔗 Aleyda Solis’s LinkedIn Post
Graphite x Axios study (May 2026), based on 55,400 articles randomly selected from Common Crawl and processed by three AI detectors (error rate below 2%).
After the launch of ChatGPT, predominantly AI-generated content exploded (36% in 12 months, 48% in 24 months), reaching approximately 50% of all content published on the web.
However, since Q1 2025, this has stagnated: 50% AI / 50% human, a figure that has remained stable for five quarters.
The explanation: there’s a barrier at the end. Only 14% of articles displayed in Google are AI-generated—and only 7% of those in position 1.
Don’t be mesmerized by the 50%. The real figure is 7%. The fact that half the web is AI-generated is irrelevant if that half almost never appears in the answers that matter. The panel is saying something simple: everyone has tried flooding the web with AI content, and they’ve all been left with nothing but tears when they see the performance of this “soup” content. So what do we do? We stop driving by volume. The real question is no longer “how many articles this quarter,” but “what can I publish that no one else can publish”: your data, your expertise, your point of view. Less, but inimitable. Then “cuttable” and “reformattable” (in English, “repurposable”) as needed, to share this content wherever it’s relevant to your audience. That’s the pivot.
Tomek Rudzki ( Peec AI) has Analyzed 500,000 prompts: while we’re talking about ChatGPT and Perplexity, it’s Google that’s in control the essentials of AI visibility.
2.5 billion monthly users for AI Overviews (announced at Google I/O 2026) — more than ChatGPT, Claude, and Perplexity combined.
AI Overviews appear in 87% of searches (frequency increased from 56.9% to 86.7% in one year).
It’s not just informational: 88.5% of commercial queries at the bottom of the funnel trigger an AIO. Google therefore influences who is recommended at the point of purchase.
A significant difference: in the EU, AIOs only appear in 76% of searches… and in France, at 0%. The only major market where neither AIO nor AI Mode has been deployed to date. #despair…
While everyone’s scrutinizing ChatGPT and Perplexity, the elephant in the room is Google & AI Overviews—the world’s largest AI search platform, the one that decides who gets recommended at the point of purchase. But here’s the twist: in France, AI Overviews aren’t even deployed. Zero percent. So you’re watching the wave from the beach, while it’s already swept everything else away—and that’s a rare opportunity. Build your presence now where Google will be digging (YouTube, Reddit, your entity authority) to be ready when it hits France. Waiting for the rollout, or pretending “move along, nothing to see here, GEO is just SEO” (so many people still say that…), will already be playing catch-up.
🔗 AI Overviews is the most undertracked AI search: 500,000 prompts show why
We already decoded it in a previous edition—but a recent study has reignited the debate about its actual usefulness for AI. So, let’s revisit it.
→ Quick recap: Schema Markup is a language of invisible tags (often in JSON-LD format) inserted into a page’s code to tell machines, in black and white: “this is a price,” “this is a customer review,” “this is the author.” In short, it translates your page into data that search engines understand without having to guess.
→ What just happened: Ahrefs tracked 1,885 pages that added schema markup, compared to 4,000 control pages. Verdict: it made virtually no difference to AI citations (AI Mode +2.4%, ChatGPT +2.2%—statistical noise; AI Overviews even -4.6%). Immediately, half of LinkedIn cried victory: “Schema was a scam, GEO was hot air.” Barry Schwartz, however, reported the study without getting carried away.
→ What the study actually says: And here, a critical expert on the article sets the right boundaries:
The tested pages were already heavily cited (100+ citations before the schema was added). Ahrefs therefore measured whether the schema boosts the ranking of pages that the AI already loves: no. But it never tested whether it helps a page get on the AI’s radar — the real question.
In the same study, however, Ahrefs found that pages cited by the AI had 3x more schema markup than others. The critic puts it well: this isn’t a contradiction, it’s an indicator. The schema has an effect elsewhere.
Where, exactly? Not when the AI writes its response (at that moment, the top 5 AIs read the visible HTML, not your schema — SearchVIU test), but earlier, when the search engines build their index. As evidence: according to him, Bing has acknowledged running an indexing pipeline that consumes structured data before AI intervenes, and Microsoft explicitly recommends schema markup for inclusion in AI search.
→ The analogy: The schema markup is like the bouncer at the nightclub entrance, not the DJ. It decides who gets in (which pages enter the AI’s radar). Once inside, dancing, putting on a nicer jacket doesn’t make you more popular. Ahrefs tested people already inside—naturally, the jacket made no difference. This doesn’t prove the bouncer is useless: they’re the ones who let you in.
Stop expecting Schema to “boost” your citations on already visible pages: on this specific point, the study is sound, it’s a losing bet.
Keep Schema as an infrastructure component, within a clean technical bundle (readable HTML, server-side rendering, recent update date, access granted to search engine crawlers, etc.) designed to be found and indexed. Removing it “because a study says it’s dead” would be the opposite mistake.
Beware of definitive claim: “Schema is dead” and “Schema is magic” are both false.
The real news isn’t “Schema is dead”—that’s a category mistake, and the study’s most critical post names it best. Ahrefs proved one precise and useful thing: Schema doesn’t amplify citations from pages that AI already favors. But Schema never played at that level—it plays upstream, at the moment it enters the radar. Neither overhyped turbo nor dead gimmick: it’s an entry ticket. And you don’t throw away your ticket just because it’s no longer useful once you’re already in the room.
🔗 We Tracked 1,885 Pages Adding Schema. AI Quotes Barely Moved
🔗 Study: Adding Schema Did Not Improve AI Citations On Google, ChatGPT & More
That’s it, Digital PR (almost) got its moment in the spotlight at Google I/O. Fery Kaszoni tells us he gave a talk on the subject there — and the Googlers, all excited, apparently started imagining their own PR stunts right there in the middle of the event. For real.
Actually, no… it’s a big joke and a really cool fake AI that made a lot of people laugh. But since every joke has a good grain of truth to it, I’ll slip it in as an aside in this “SEO” section.
Welcome to the section where we optimize for the whole world, not just for the search bar 😉
Moz delivers a blunt diagnosis: when PR, SEO, and content operate in silos, efforts are duplicated, KPIs are missed, and the brand message becomes fragmented across channels.
Their example is a perfect one—the 2017 Pepsi x Kendall Jenner ad: not just a creative blunder, but a complete lack of alignment between the ad, PR responses, and on-site content when the crisis hit.
On the other hand, when integrated, the two disciplines feed off each other: PR generates awareness (media, influencers, storytelling), SEO keeps this content discoverable long after the news cycle, and Earned Media gains a second life in SERPs and AI Overviews.
And Moz makes it clear: LLMs draw from social media, SERPs, and Earned Media—PR and SEO no longer exist in separate ecosystems. Continuing to compartmentalize them means giving AI a fragmented image of your brand. The goal is no longer to be mentioned, but to be discovered, remembered and chosen.
Everyone’s been saying “break down the silos” for ten years (I, for one, have been shouting about it for 15, ever since my book on content marketing and before that, my online academy on the subject), and the silos are still there. Because it’s not a matter of goodwill: “hold meetings, align your KPIs” doesn’t solve anything.
It’s a problem of organizational structure and budgets—as long as PR and SEO depend on two managers, two budget lines, and two sets of objectives, they’ll remain as separate as Mars and Venus. The real lever isn’t another meeting; it’s an executive committee decision: a single person responsible for visibility, from start to finish, with control over both budgets. Without that, AI will continue to read a brand that’s talking to itself in a jumbled mess.
🔗 Post by Moz — “How to Integrate PR & SEO for Maximum Brand Visibility”
Ann Smarty shares a striking observation: the Reddit AMAs (“Ask Me Anything”) she organizes in her clients’ subreddits rank very well in Google for branded searches—including ultra-competitive queries about well-known industry figures (she cites “Lily Ray”). In other words, community content is no longer content to exist in isolation: it’s capturing a place in the SERPs for your own name. And when you consider that Reddit has become one of the most cited sources by AI (see the Aleyda Solís study) and one of the top destinations after a search, this is no longer a minor detail: it’s a channel where your reputation… is either built—or hijacked by others.
For a long time, your brand page on Google was your private domain—the only search query you truly controlled. That’s over. While a Reddit AMA might rank for your name (or a competitor’s), your reputation is now negotiated in spaces you don’t own. This isn’t a reason to panic, but rather a reason to be the first to seize the opportunity: occupy the online community space before someone else does. This can be achieved either by becoming highly active in targeted communities (ultimately, it always comes back to the good old adage, “know your audience and identify where they spend their time online”); or by creating your own.
Brendan Hufford analyzed his 332 posts from the year.
The numbers tell the story: an average of 16,500 impressions for text, compared to 7,800 for video.
Since LinkedIn replaced its algorithm with an AI system in March—which measures time spent, saves, and comment depth—video has fallen even further (views down 36% year-over-year), while text, carousels, and images have climbed.
But the real blow comes from AEO: LinkedIn has become one of the most frequently cited domains by AI search engines for professional queries… except that video is neither indexed nor cited by ChatGPT or Perplexity. A LinkedIn video is therefore useless both on and off the platform.
Beware of the misconception that “ video is dead.” This month, other studies show the opposite: YouTube video is THE source that correlates most strongly with AI visibility. The whole difference is in the container . A YouTube video is transcribed, indexed, backed by Google — the AI reads it; a video Being locked inside LinkedIn is like a walled garden she doesn’t know Don’t read. The lesson is not So not “ stop the video ”, it’s “ stop of it produce there where the machines are blind people ”. Films for YouTube, writes for LinkedIn.
Jason Feifer, editor-in-chief of Entrepreneurs Magazines, received confirmation directly from LinkedIn (via Laura Lorenzetti): the platform is rolling out new systems to curb three major problems—generic AI-generated posts and comments, tools that automatically spam comments, and attention-bait videos (random images and misleading, catch-all business advice).
Low-quality AI content will be algorithmically limited: it will no longer be allowed to appear beyond your first-degree connections.
The reason? Content creation has surged by 14% year-over-year, the feed is saturated, and the AI mush is stifling good content. LinkedIn hopes that by reducing the slop, original posts will get some breathing room. Significant changes are expected in the coming months.
It’s the same story as the Graphite study (↑), viewed from a different perspective. Google already filters the AI mush in its index; LinkedIn is curbing it in turn. The message from the platforms is becoming unanimous: volume is no longer an advantage, it’s a liability. Good news for those with a real voice (which you can actually do on LinkedIn) — less generic noise, more space for originality. The dividing line is no longer “AI or no AI,” it’s “generic or unique”: use AI to get faster on what only you can say, never to produce the same thing or copy others.
And the article he quotes LinkedIn is fighting back against AI slop and AI Comments
Darren Shaw points out a counter-trend: last week, DuckDuckGo announced a 76% increase in app installs in the United States compared to normal.
The reason? : Internet users no longer want AI to be the default in every search. Sometimes, you just type “best headphones 2026” without wanting Google to summarize the entire web before you’ve even clicked—you want to compare real websites, read reviews, and decide for yourself. DuckDuckGo’s argument isn’t “zero AI,” it’s “AI as an option, not imposed.” Darren himself clarifies: Google remains king; only a fringe will switch. But the signal is there—people want choice.
This is the dissenting voice of this edition—and it’s a healthy one to hear. While everyone is screaming “AI everywhere, adapt,” a segment of internet users is doing the opposite: they’re fleeing imposed AI. So don’t assume that 100% of your audience wants the answer spoon-fed by the machine; a real segment still wants the blue links, the comparison, the freedom to choose. And remember the pattern: in a race where everyone is forcing AI by default, “giving choice” becomes a selling point—just like “ad-free” was for ChatGPT not long ago. One person’s constraint becomes another’s selling point.
This book puts everything into perspective: you can rank highly on Google and yet remain virtually absent from the data that trains AI. The culprit? CCBot—the Common Crawl bot from which many models derive their URLs—is one of the most blocked user agents on the web. Thousands of sites unknowingly sabotage themselves by relying on their SEO or RAG (Real Aggregate Analytics) to appear after the fact, instead of being aware of the model from the outset. Stephen Burns’ (Common Crawl Foundation) guide provides a concrete audit, which can be completed in about 90 minutes: CCBot access, presence in the index, Harmonic Centrality, completeness of structured data, and server-side rendering.
The shift in focus is abrupt—but so true: the real question is no longer “am I visible in the results,” but “am I in the brain of the model?” This needs to be audited this week.
Cloudflare (which sees traffic from 25% of websites and 45% of the top 100,000) has released a staggering figure: for the first time, bots account for the majority of global web searches—57.3% compared to 42.7% for humans. CEO Matthew Prince predicted this shift for 2027: it’s already here. And here’s the crucial detail, reported by Corey Northcut via Cloudflare Radar:
Training LLMs has become the primary role of bots.
Google’s share of all bot activity has dropped from 70% to 40% in just one year.
Meta and Claude now crawl more content than ChatGPT.
The shift is also behavioral: a human making a purchase visits 5 websites; an AI agent can visit thousands—very real traffic and server load, but without clicks, without ads seen, without customer interaction. As Northcut summarizes: we spend our time scrutinizing outputs (referral traffic, citations, active users), while Cloudflare finally reveals the inputs—what machines ingest from you, upstream.
This is the flip side of the previous entry (↑ Rand Fishkin): Rand describes a Google that wants to be the web; Cloudflare shows the bots that are already ingesting this web to train themselves. The inputs and the strategic consequence are mutually reinforcing.
This is the blind spot in all your dashboards. You’re measuring who quotes you and who clicks—outputs—while training bots (Meta, Claude, and increasingly less ChatGPT) are consuming your content right now to shape, for years to come, what AI “knows” about you. The real playing field is shifting: ingestion, not citation.
→ Two key takeaways for your executive committee meeting: your website is now as much an API for machines as it is a showcase for humans, and your AI visibility is no longer judged solely by clicks received but by what’s consumed beforehand. Start by taking a look—Cloudflare Radar is free.
🔗 Search Engine Land article : Cloudflare: Bots now make up 57% of webpage requests
🔗 Corey’s post (and his exchange with Andy) on LinkedIn (click on the image)
Josh Blyskal analyzed approximately 7 million recent citations and ChatGPT’s “query fanouts” (the intermediate queries it formulates before responding).
→ The result: ChatGPT explicitly requests Reddit 24 times more often than at the beginning of the year, and Reddit has once again become its number one source (8.5% of all its citations). But the real takeaway isn’t Reddit—it’s that these fanouts reveal the AI’s source preferences based on intent:
“reddit” → it’s looking for firsthand experience.
G2”, “Capterra” → third-party reviews.
“GitHub”, “docs” → proof of implementation.
“pricing”, “alternatives”, “vs” → purchase decision pages, not your generic category page.
For certain queries, ChatGPT has already decided where the answer will come from—and no amount of new 1,500-word SEO pages will change that. The question is no longer “Is my content good?” but “Am I present where the AI has decided to search?”.
As Blyskal summarizes, citations tell you who got the answer; fanouts tell you where the answer was allowed to come from.
This might be the most important piece of information this month for your visibility strategy. Until now, we optimized our content hoping to be chosen; now, we’re discovering that AI pre-selects the type of source before even reading anyone’s work. You can write the best page in the world: if ChatGPT has decided that this query “belongs” to Reddit or G2 and you’re not there, you don’t exist. AI visibility is now earned upstream—by being present in the right places for each search intent—not downstream, on your own page.
🔗 Josh Blyskal’s LinkedIn Post
Cyrus Shepard decodes what Google truly rewards, based on patents, evidence from the antitrust trial, and the API leak. Three behavioral signals keep recurring:
badClick: The user clicks, isn’t satisfied, and immediately returns to Google. Bad sign.
goodClick: They stay, read, and engage. Good sign.
lastLongestClick: They get their complete answer and never click anywhere else. The holy grail.
And the stakes go beyond traditional SEO: these same signals feed AI responses (Google “anchors” its AI Overviews to 70 days of search logs, and ChatGPT relies heavily on Google results). The strategic consequence: stop thinking “keyword,” think “journey.” Someone searching for “movie showtimes” actually wants reviews, cast information, the theater—and ultimately, their ticket. If you force them to return to Google to complete their search, you’ve lost. The winning page answers the query, anticipates follow-up questions (obvious and not obvious), and moves the user to the end.
Here’s a strategic compass of rare simplicity (frankly, I love it!!): don’t aim to rank, aim to be the last click. And that’s excellent news, because this signal is virtually impossible to fake—Google remembers 13 months of interactions, bots can’t do anything about it. For once, the algorithm rewards exactly what you should want to build anyway: a page so comprehensive that the user no longer needs to look elsewhere. And this ties in with what we said earlier (see the Graphite study): being the last click means being the page that says what no one else is saying.
And if you read this in conjunction with Kevin Indig’s study (see the data section), it shows what is now a constant: that the SERP journey has become non-linear—we read, we scroll up, we compare. Cyrus provides the solution: to be the page that closes this journey, so that the user has no further reason to come back and search.
🔗 Check Cyrus’ awesome Substack: Zyppy Signal with Cyrus Shepard
Rand Fishkin wrote the most shared article of the month.
His thesis: For 25 years, Google said, “Make good content,” and that was enough.
Except that’s over. Google no longer wants to be the gateway to the web—it wants to be the web: it indexes your content, transforms it into a commodity in its AI interface, and only sends you clicks if you pay (ads) or offer something it can’t provide itself. Rand’s striking statement:
“Influence is the new traffic; engagement, the new top of the funnel; brand search, the middle; sales, the only true measure.”
His solution: Stop focusing on content, build inimitable products—a physical good, a service, an experience that AI can neither invent nor copy (it will never forge a chef’s knife, tailor your suit, or cook the meal of your life). Because an inimitable product is a defensible moat; inimitable content, on the other hand, is not.
But… you have to read the post to see that Robert Rose offers a third way.
From Robert’s standpoint, the holy grail of the statement “content is a commodity” is only true if we accept Google’s definition: content = data. But nobody ever subscribed to Stratechery for facts, nor listened to Bourdain for restaurant recommendations. The asset is the human element, taste, and trust built over time.
→ His third approach (Robert’s): the inimitable medium—content inseparable from the person who creates it. His punchline: AI can replicate the latest thing, but not the relationship that makes you believe in the next one.
→ Rand’s response: Stratechery is inimitable… but it’s a content product, whose value is declining—and today, you wouldn’t build such a brand through a blog. We’re swimming against the current, not with it.
Rand and Robert aren’t really opposing each other—they’re describing the same shift from two different perspectives. The uncomfortable truth is shared: content is no longer the asset; it has become the marketing of the asset. The question remains: what is your asset? And can your (only) asset be content? If it’s a product that AI can’t manufacture, you have a moat (Rand); if it’s a relationship of trust that it can’t earn, you have one too (Robert). And for us, creators and media brands, this is where it all comes down to: not being the content, but being the inimitable thing and using content to get the word out. The real test for an executive committee: if your only asset is content that AI can “market,” you don’t have a moat—you have a countdown (sorry to be blunt… but I mean it).
🔗 Inimitable Product is the New “Make Great Content” - SparkToro
Optimized website + Google Business Profile + around fifty citations + a few reviews: this is now the bare minimum.
To be recommended by AI in local search results, you need to control what the entire web is saying about you. Local SEO has become a word-of-mouth system that LLMs (Local Marketing Links) read.
Spy on the AI: Run your branded queries about twenty times in an LLM (the results change each time) to see who it recommends and what sources it cites. Then get yourself mentioned on those sources: guest post, podcast segment, YouTube. (Tools: Waikay, Gumshoe.)
Diversify your reviews: Not just Google: Yelp, BBB, Facebook, platforms in your industry. And respond to all your reviews—the AI reads the responses too.
Frame your review requests: Guide the customer toward what the AI will ask: what problem was solved, punctuality, value for money. A detailed review is worth ten “great service!”
Be everywhere: Reddit, forums, LinkedIn, hyperlocal press and blogs, your city’s best-of lists, press releases. The AI scours even the most obscure mentions.
Write for the machine: The answer at the top of the page (LLM scans the beginning), then structure the information into three semantic categories: drop the “we,” write “[Brand] is [a plumber in Denver],” “[Brand] offers [drain cleaning].” And bring something new—your real-world experience, not just the same old industry jargon.
People often think Google Business Profile is only for restaurants and plumbers. Wrong. Even for consultants and solopreneurs, it’s one of the few sources Google directly feeds to its AI when someone types “consultant near me.” The only requirement: seeing your clients in person, even if it’s just while traveling—you can then hide your address and only show your service area. 100% remote, Zoom-only? To get it, you need a video of yourself in your offices, which must be at the address you provide. If you don’t qualify: focus all your efforts on your website, its Schema, and of course, all the platforms where it makes sense for your audience to find you, and above all, never create a fake address; it’s an immediate suspension.
🔗 Darren Shaw, Search Engine Land — The new playbook for localized AI search optimization
Cyrus Shepard answers the real question → which sites will keep traffic when AI keeps the clicks?
Google Zero = the point at which Google stops sending you traffic. For some publishers, this has already happened (almost zero visits), and the default AI Mode will accelerate this trend. An important distinction: Google won’t cut everything off—people are still searching for sites, and they’re actively seeking them out.
Danny Sullivan (Google) shared a slide contrasting “commodity” content—the kind anyone can produce, which Google no longer wants to rank—with “non-commodity” content.
Shepard analyzed hundreds of winning and losing sites and identified 17 types of content that survive. Their unique commonality: they are first-party sources. The data comes from you. No one else can produce it.
The best bets → original search, communities (where people share their perspectives), and transactional product pages. The owner, never the associate.
For ten years, we optimized for volume → publish more to rank higher. Google Zero has just made this reflex a bit suicidal. If a machine can write your content, it’s because a machine has already written it—better, faster, for free. The only remaining gap is what it can’t invent: your data, your experiences, your opinions. Stop feeding AI what it already knows. Give it what it doesn’t yet know.
🔗 Cyrus Shepard — Google Zero ( LinkedIn video )
Ethan Smith (Graphite), based on the podcast The Answer Engine (Webflow) with Guy Yalif.
Today, “best tool for X” lists and affiliate marketing dominate AI citations for “best tool for X” queries—often biased and with little informative value. This exploits early SEO vulnerabilities: it works until the platforms shut down. Don’t build anything on it.
What lasts: Product content (features, use cases), comparison pages, and above all, content backed by proprietary data. The unique wins.
The number one lever: Answer the questions your customers actually ask—not the ones you imagine they’re asking.
Real-world FAQs. One per page, fueled by your sales calls, customer service emails, and customer feedback. Not by an internal brainstorming session.
Glossaries: only useful if they go beyond the definition to cover features, use cases, and integrations. Otherwise, they have no effect.
Measurement: embrace the ambiguity. AI traffic is probabilistic (retrying each request multiple times), varies depending on whether the user is logged in or not, and remains overwhelmingly zero-click. Follow the LLM benchmark in your analytics, then add a dedicated tool for citation share.
Smith is fundamentally right (like often) : building your AI visibility on “best of” lists and affiliate marketing is simply rehashing old SEO weaknesses—it works until the day it doesn’t, in this case, when the platform shuts you down. The real defense is content that can’t be copied: your data, your real customer case studies. But beware of dogmatism: the Landwehr study shows that this “weakness” now accounts for up to 17 points of visibility. So, you have to play both sides—secure your spot in lists while it pays off, and build something truly unique for the future.
🔗 Ethan Smith (Graphite) — The Answer Engine, ep . 1
A study of over 5 million data points quantifies the impact of third-party rankings (Top 10 lists, comparisons) on LLM recommendations.
The brand ranked #1 in a frequently cited listicle gains 13 to 17 visibility points (probability of being mentioned) and moves up 1 to 2 places in the response.
Brands in positions 2 through 10 also benefit: +6 to 16 points, and a slightly better ranking.
The effect is most pronounced in new markets, where the model has virtually no pre-loaded knowledge: it relies entirely on the sources it seeks out. In mature markets, these sources often simply confirm what it already “knows.”
The real surprise: this only works with third-party lists. Your own rankings and those of your competitors have almost no measurable effect.
The verdict is brutal - liberating: you don’t gain AI visibility by talking about yourself, but by being ranked number one by someone else. Your “why we’re the best” page doesn’t count—that’s what you say. The real GEO budget line, therefore, is to secure first place in the rankings already mentioned by LLMs, especially in a new market where AI has nothing else to offer. Ethan Smith’s warning remains: this lever resembles an early SEO flaw, and the platforms will eventually fix it. So exploit it quickly—but don’t make it your entire strategy.
🔗 Malte Landwehr et al. — SSRN study “Cited-Listicle Rank-Tier Exposure…”
While searching through ChatGPT’s citations, Chris Long stumbled upon an “AI Information” page from the agency Seer Interactive (Wil Reynolds, Alisa Scharf)... which the model copied verbatim. He then replicated it for his own agency:
A page in pure Markdown, linked from the footer, with a section that clearly told ChatGPT what it should remember about the brand.
The tricky test: he slipped in “we work with companies with over $30M in ARR,” information that appeared nowhere else on the site.
The result: in less than 48 hours, ChatGPT cited the page and often used the exact wording, including the ARR.
Take this with a grain of salt: it’s a single, self-reported experiment, and the effect isn’t systematic (Long himself anticipates being accused of contextual bias). But the signal warrants your own testing. And a word of caution: what you write on these pages, AI can regurgitate word for word—true or false. Handle this ethically before the platforms put a stop to it.
🔗 Chris Long ( Nectiv ) — “AI Instructions” experiment
The skeptics’ position is clear: everyone is selling GEO right now, and a good portion of it is a sleight of hand. The culprits are self-promotional article lists, and especially the hidden instructions slipped into “Summarize with AI” buttons to trick the model into treating a brand as a reliable source—a manipulation Microsoft has dubbed “recommendation poisoning.” The Verge ran a well-researched investigation on the subject.
Britney Muller’s verdict is scathing: those who promise to get your brand cited by AI are selling a guarantee they can’t deliver. What actually works is much harder to package into a commercial offer—real authority, genuine third-party mentions, and content robust enough to withstand a machine reading the entire page, not just the headline. Nothing new: that’s what good content has always required, long before it was given an acronym.
🔗 Intercept’s LinkedIn post quoting Britney
Google has just published its official guide to optimizing for AI search, and its position is unambiguous. SEO remains valid because its AI features (AI Preview, AI Mode) run on its usual ranking systems.
And on the contentious issue, the verdict is in: from Google’s perspective, optimizing for generative search is equivalent to optimizing the search experience—in other words, SEO. AEO, GEO: two acronyms for a discipline that already exists…
…according to Google!
Better yet: Google provides a black-and-white list of the “tricks” its search engine can ignore—llms.txt files and other special markup, breaking content into small chunks, rewriting for AI, hunting for artificial “mentions,” and over-optimizing structured data. For Google, none of this changes anything.
The referee blows the final whistle—in his favor, of course. Google relegating GEO to the “SEO, business as usual” category, WOW… talk about a scoop!! Is it really so surprising that Google wants us to optimize our content… for them? 🤔
But then again, when the one speaking holds 85% of the online search market share, you can’t help but think that following their recommendations isn’t such a bad idea.
That being said… I’m going to offer my Solomon-like judgment: Google only talks about Google! As soon as you target ChatGPT, Perplexity, or models that learn via Common Crawl (↑ reread this month’s article), the rules change—rendering without JavaScript, having a presence on Reddit, and files that Google deems useless become real levers again. “GEO is SEO” is only true in one respect. Except that in 2026, the rest of the championship is also played elsewhere.
🔗 Google Search Central — AI Optimization Guide generative
Why it’s so useful:
Cloudflare routes roughly a quarter of the web’s traffic, so it sees, almost in real time, how AI bots behave on a global scale. While all your other tools measure outputs (citations, incoming traffic), Radar shows you the inputs—who’s crawling what, why, and who’s sending you back traffic in return.
THIS IS HUGE - It’s the only public window into this layer (↑ we explored its full strategic implications in the Strategy section).
How to use it:
Monitor the AI bots that are rising (and falling) in your sector: you’ll know who’s interested in your content before anyone else.
Filter by “purpose”: training / search / user action. Today, training dominates (~80% of AI crawling).
Look at the “crawl-to-refer ratio”: how much each AI takes versus how much it returns to you. The numbers are striking—ClaudeBot crawls tens of thousands of pages for every ~1 return visit. Enough to put those fantasies about “AI traffic” to rest.
Identify the adoption of robots.txt and new “agent” standards: useful for deciding what you allow (or not) to ingest.
If you’re a Cloudflare customer: compare your own site to the macro—are you crawled more or less than your peers, and who’s actually sending you traffic?
🔗 radar.cloudflare.com/ai-insights
Why it’s amazing:
Remember the tactic “if you don’t use English, AI doesn’t know you as well: focus on your brand signals” (↑ GEO Tactics Pulse). This tool is a direct implementation of that tactic. It generates an HTML page (with schema markup) that you publish on your website and link to from your footer. Its sole purpose is to provide AI with clear, complete, and 100% text-based facts about your company.
This is “AI-first” content: your website is the primary training source for your brand—so it’s best to give it a clean profile rather than leaving it to guess. (Orbit’s killer reminder: AI doesn’t read your images or pricing logos, only text. If you want it to know this, write it down.)
How to use it:
Give it everything. Your services, the sectors you target, the details that come up in sales meetings, related services, even special cases.
Expand the descriptions, be exhaustive—not marketing-driven. Here, completeness trumps the punchline.
Retrieve the generated HTML page (with its schema), publish it, and link to it from your footer.
For your French-speaking audience, it’s even more profitable: AI has less data about you in French, and a clean practice page fills that gap perfectly (↑ see the 0% AIO in France and the lack of French sources mentioned above).
🔗 AI Training Page Generator — custom GPT Orbit Media + “AI-Friendly Websites” guide
→ Not a tool, a prompt. Free, to paste into Google Sheets.
Why it’s great:
You already have thousands of keywords with real search volume (Ahrefs, Semrush, Similarweb). This trick recycles them into a single column: the way a real person would ask the same question to ChatGPT, Gemini, or a voice assistant—in a complete sentence, preserving the exact intent. In two minutes, you go from a list of keywords to a map of real conversational queries. It’s the fastest bridge between your SEO of yesterday and AI search of today (↑ compare this to Josh Blyskal’s “fanouts” I shared earlier, to understand how people actually formulate their requests).
How to use it:
In Google Sheets, paste your keywords into column A, then use the =AI() (or =Gemini()) function with Lily Ray’s prompt, pointing to your cell (A2). You can have it adjusted by an LLM to fit your industry.
🔗 Lily Ray’s LinkedIn post for you to copy paste the prompt
(↑ Yes, the same as in the data section). Paid service (starting at $80/month), 30-day free trial, France 🇫🇷 supported (YAY!!).
Why it’s good:
Finchling starts from a premise that’s exactly the same as yours: branding has become one of the most powerful signals in search, and AI makes it even more critical. In other words, the tool scans thousands of sources and tells you not “who mentioned you” (like traditional keyword monitoring) but “which news story you have a PR angle to leverage, and why.”
Mordy Oberstein, on the other hand, uses it for a more ingenious reason: the “Trending PR Campaigns” library becomes a strategic intelligence platform—identifying positioning gaps and observing how media narratives shape brand perception, including in LLMs. His key insight: a discreet and subtle campaign can have a more lasting impact than a big, noisy one.
How to use it:
Create your watchlist: your brand, your competitors, the topics you want to “own.”
Receive curated alerts with a ready-to-use angle + the “why it matters” + data-driven angles to pitch (the story only you can tell).
Activate “Plan Ahead” to map upcoming moments (up to one year) and plan your campaigns instead of reacting to the news.
Monitor competitor campaigns that are breaking through—and learn from them.
Filter by region (France) and explore the “Trending Campaigns” library to see which brand narratives are performing well.
👋 That’s all for this month! Thanks for reading. I’ll see you again in July in the same format. If you’ve been forwarded this newsletter, you can subscribe here👇😘.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.