Hi everyone!

This week a really good study landed that shows the goal everyone's been chasing in AI search is actually only half the goal. We all keep asking, "Is ChatGPT citing me? Is Gemini mentioning me?" Burson and Profound came along and said: you're asking the wrong question. The real issue isn't being visible, it's being believed once you're visible.

Think about it: AI mentioned your brand, but nobody believed the sentence. Or worse, everyone believed it when it mentioned your competitor, and shrugged when it mentioned you. That's exactly the "credibility paradox" we're talking about this week.

This actually picks up right where we left off last week. If you remember, last week we built a Claude task together that audits your AI visibility every week. Back then I said measurement isn't a one-time snapshot, it's a trend you have to track, and the question of "how much original value did you add" matters at least as much as "did you show up." This week's Burson study crowned that sentence with numbers: being visible is one thing, being believed is something else entirely.

On top of that, Google didn't sit still this week either: it officially added "optimizing for generative AI" to its SEO docs, said clearly that it doesn't endorse third-party tools, and published a "myth-busting" list declaring that many GEO tactics like llms.txt and chunking are unnecessary for its own AI features. It also reminded everyone, again, that buying citations now counts as spam.

Grab your coffee, let's dig in.

Burson, in its "The Credibility Paradox" report published on June 2, 2026, surfaced a striking finding: a brand showing up in AI answers doesn't mean that answer will be believed. The research, run with AI analysis platform Profound, examined 85 companies across seven major AI answer engines and produced more than 55,000 believability forecasts.

The conclusion: visibility and believability are two separate things. Your brand can appear in AI Overviews or in ChatGPT, but what's said may not land as convincing. This moves the GEO (Generative Engine Optimization) conversation from "how do I show up" to "how do I become believable once I show up."

Burson, using its own Decipher tool (built with cognitive AI firm Limbik), scored AI answers in the eyes of three separate audiences: the general public, opinion elites, and business decision-makers. 85 companies were evaluated across eight different reputation dimensions.

Two findings stand out. First, believability shifts significantly by audience. Business decision-makers find AI answers 10% more convincing than the general public does. So the same answer carries different weight depending on who it's shown to. Second, the type of content determines believability. Fact-based claims about a company's innovation, products, and workplace culture are found far more believable than claims about more subjective qualities like leadership and governance.

This study takes the GEO conversation up a level. For the past two years we've chased the same metric: mentions and citations. Those are visibility metrics. Burson is saying they're necessary but not sufficient. Because AI mentioning you doesn't mean the user believed the mention.

The fact that fact-based claims are found more believable than subjective ones is a direct roadmap for us. In AI search, subjective sentences like "we're the industry leader" float off into nothing. Concrete, verifiable claims like "we brought this feature to this product, and got this result on this date" stick. In other words, believability comes from evidence.

This study confirmed something I've been saying in this newsletter for years, with numbers, and I loved it. I always say AI doesn't love generic (commodity) content, it loves evidence-backed content. Burson went one step further and measured this: evidence-backed content isn't just cited more, it's also believed more.

I've seen this play out again and again in client projects. Two brands show up side by side in AI. One is full of self-appointed adjectives like "leading solution provider," the other speaks with hard numbers, real cases, and verifiable claims. The second brand stands firmer every single time. Burson's study explains why: facts convince, adjectives don't.

For me the key takeaway is this: from now on, when we run a GEO audit, we won't just look at "did I show up." We'll also look at "when I show up, what's being said, and is it believable?" That's a new dimension added to our measurement set.

Do this this week: ask ChatGPT or Gemini a question about your brand and read the answer word by word. Don't just check "is my name there," ask "would I believe this sentence?" Is the answer full of subjective adjectives ("reliable," "leading," "high-quality"), or does it speak with concrete evidence (real numbers, dates, product specs)? If it's subjective, back those claims with facts on your website. AI only carries evidence into its answer if it finds that evidence on your site.

Speaking of believability, I can't get through this week without telling you a little story that happened to me. Because it made me live, firsthand, the AI-world truth that "visibility is nice, but control sits elsewhere."

This week on my YouTube channel I shot a video showing what you can do with Anthropic's new model, Claude Fable 5. Fable 5 is Anthropic's new model, made public on June 9, especially strong at vision and software engineering. In the video I ran a few design experiments with the model, and the results were genuinely impressive.

Then guess what happened? Shortly after I published the video, the US government shut Fable 5 down. Apparently they saw the design I made and said, "this is too much, we need to stop this." Joking aside, the real story is far more interesting than you'd think.

The US government forced Anthropic to shut down its two most powerful models at once, Fable 5 and its safety-loosened cousin Mythos 5, worldwide. The stated reason: national security. But the details are odd. The order was officially issued only for "foreign nationals"; yet because it's impossible to tell in real time which user is a foreign national, Anthropic had no choice but to shut the models down for everyone. The company's other models aren't affected, only these two are disabled.

The claim is this: a method was found to bypass the model's safety locks, a jailbreak. According to Anthropic, that "dangerous method" amounts to having the model read a code file and fix the bugs inside it. What's more, other models on the market can already do the same thing. Anthropic complied with the order and cut off access, but it also stated its objection clearly: if this standard were applied across the whole industry, no new model could ever launch.

My take: The lesson here is even bigger than this week's headline. For months we've been discussing the limits of AI in terms of compute, AGI, and data. Yet the first hard wall came from somewhere we never expected: regulation. And it broke just weeks before Anthropic's IPO; the timing is pretty unfortunate for the company. For Chinese models, on the other hand, it cracked open a door the other way: as the US blocks its own model, the gap with its rivals narrows. Anthropic sees this not as a permanent obstacle but as a misunderstanding, and is working to restore access.

For me the real message is this: AI models are not under your control. A model is here today and can be shut down three days later by a government decision. Just like AI search visibility: it cites you today, and goes quiet tomorrow when an algorithm or a regulation shifts. So don't stay dependent on a single model, a single platform, a single channel. I shot the video with Fable 5, the model got shut down; but the method, the logic, and the content I shared are mine, and they're still standing. That's what lasts: tools come and go, your original contribution stays. Anyone who wants to watch the video can find it on my channel, and the method works just fine with other models too.

What happened? Google updated its Search Central docs on June 5, listing "optimizing for generative AI" as a legitimate job an SEO can do. This came on top of the "Optimizing your website for generative AI features on Google Search" guide published on May 15. In the same update Google clarified two more critical things: that it doesn't endorse third-party SEO/GEO tools, and that buying citations, manufacturing fake mentions, and producing scaled content to manipulate AI answers all fall under its spam policies.

What it means for SEO/GEO: This is the moment GEO became official. For years people argued whether "optimizing for AI" was a real discipline or something consultants made up. Google now says, in its own docs, "yes, this is legitimate work." But in the same breath it closes the shortcuts: what's legitimate is optimization, not manipulation. The era of buying citations is officially over.

My take: Don't mix up two separate stories here. The first is good news: Google accepted GEO as legitimate, which adds value to our work. The second is a warning: saying "we don't endorse third-party tools" is a shot across the bow at the "we'll get you into ChatGPT" magic-box vendors out there. I've been skeptical of them from the start, because AI visibility isn't bought from a tool, it's earned with content. Google has now landed on the same line.

Practical step: If a GEO tool or agency is promising you "guaranteed ChatGPT visibility," question that promise this week. Google's official line is clear: earned visibility lasts, bought visibility is a risk.

What happened? Inside that same documentation update, Google also published a "Mythbusting generative AI search" section and flagged, one by one, many of the tactics sold in the name of GEO over the past two years as "unnecessary." Google says plainly: to show up in AI search you don't need an llms.txt file, you don't need to break content into small pieces (content chunking), you don't need AI-specific rewriting or special schema markup. It even defines AEO and GEO and adds: "From Google Search's perspective, optimizing for generative AI is optimizing for the search experience; so it's still SEO."

What it means for SEO/GEO: This is an important clarification that cuts through the noise around GEO. There's a flood of advice out there saying "rewrite your site for AI," "add an llms.txt," "chunk your content." Google is saying most of that isn't needed for its own AI features. The core hasn't changed: quality, original, well-structured content. So it's not about learning a new technical discipline, it's about doing the SEO you already know, well.

My take: I loved this story, because I heard Google say out loud what I've been saying for years. Most of the new titles like "AEO specialist" and "GEO specialist" are a label slapped on top of a solid SEO foundation. Instead of hunting for a separate magic file, a separate schema, a separate writing technique for AI, just make your content genuinely good. But one caveat: Google says "you don't need these," not "these are harmful." So if you want to add an llms.txt, go ahead, just don't treat it as the magic key to AI visibility. There is no magic key; there's quality content.

Practical step: If someone tells you "let's rewrite your site in this special format for AI," look at Google's myth-busting list first. Spend your budget on content quality and originality, not on AI-specific technical tricks.

What happened? Google's May 2026 core update completed on June 2 (it had started on May 21, running 11 days and 21 hours). As of this week the SERPs are starting to settle and analysts have moved on to reading the final picture. It was 2026's second core update and caused serious ranking volatility throughout the rollout.

What it means for SEO: Because the dust just settled, the real work this week is analyzing the pages that lost and won. The direction of the update matches March: sites where the content creator is also the primary source of the information won, while aggregators that summarize what others said lost.

My take: I always say don't move in a panic after a core update finishes. The first week is volatile, the second week reveals the real picture. We're right in that second week now. If you dropped, first list which pages dropped, then look for what they have in common. My observation is that the pages that fell hardest in this update were the "I wrote it, but I was really just repeating someone else's information" type.

Practical step: Open the last three weeks of traffic data in Search Console and list your 10 biggest-dropping pages. Ask each one: what's my firsthand contribution on this page? If the answer is "none," that's where the problem is.

What happened? There are two separate stories.

The good one: Google keeps rolling out its new AI performance reports that surface AI Overviews and AI Mode impressions (open to a limited set of sites for now).

The bad one: Google confirmed a 50-week data bug in Search Console, and that this data cannot be recovered. So some sites now have a permanent gap in their historical performance data.

What it means for measurement: The AI reports finally bring an official answer to "do I show up on Google's AI surfaces." But the 50-week data bug reminds us of something: never depend fully on a single platform's data. Google Search Console is a critical source, but back your data up to your own side regularly.

My take: This data bug didn't surprise me at all. I've been telling clients for years: export your Search Console data once a month and save it to your own spreadsheet. Because that data sits in Google's hands, and one day, just like this, it can vanish in a way that can't be recovered. If the data is yours, it doesn't disappear.

Practical step: This week, export the last 16 months of performance data from Search Console as a CSV and save it to your own archive. Turn it into a monthly habit.

What happened? OpenAI quietly made a big change to ChatGPT: it now embeds brand links inside the answer, as clickable links, rather than settling for footnote citations only. The result is striking: referral traffic from ChatGPT jumped 157.7% week over week, and the share of brand homepage links within OpenAI referrals shot up from 3.5% to 24% industry-wide.

What it means for AI: This is concrete proof of why AI traffic can't be ignored anymore. The volume is still well below Google; but the quality side is a different story. Visitors from ChatGPT convert at roughly 7%, far above organic search (1.76%). In other words, low volume but high value.

My take: I always say this: don't dismiss AI traffic by looking at volume, look at conversion. If a user asks ChatGPT, sees the answer, and still clicks your link, that "educated click" is very valuable. Because that user knows what they're looking for and chose you deliberately. After the branded-link update, homepage visibility exploded; that shows just how critical it is for your brand name to be recognized correctly by AI.

Practical step: Open ChatGPT and ask "what is [your brand], what does it do?" Is the answer correct, does it push you or your competitor forward, does it link to your site? If it's wrong, make your brand-basics page (who you are, what you do, who you serve) clear and current on your website.

This week the game leveled up in AI search: it's no longer just about showing up, it's about being believable once you show up, and earning it. Burson said "believability," Google said "legitimate optimization but no manipulation"; both land in the same place, evidence-backed content wins.

There's really just one sentence to this week: being visible isn't enough anymore. Burson's study, Google's new docs, even the "you don't need llms.txt" myth-busting list, all point the same way. AI and Google reward evidence-backed, believable, earned visibility; not inflated, bought visibility.

Here's the good news: this is an era where the one who does the work well wins, not the one with the biggest budget. If you'd like to keep up with these moves in real time, SEOs Diners Club is where I share the weekly read; if you want to go deeper with strategy on specific markets or verticals, that's what Stradiji does for enterprise clients in the US, UK, and Australia.

If this newsletter has been useful to you, you can add Stradiji.com as a Preferred Source in Google with this link. One click, five seconds, and we'll meet more often inside Google's AI answers.

See you next week with more.

Mert Erkal

I genuinely enjoy sharing what's moving in SEO and AI search with you every week. If you find the work useful, you can keep it going by buying me a coffee. ☕ Click 'Buy Me a Coffee' to contribute to the knowledge-sharing. Let's build more together.

Mert Erkal is a digital marketing and SEO expert with 15+ years of experience. As the founder of Stradiji, he provides consulting on SEO strategy, GEO (Generative Engine Optimization), conversion rate optimization, and AI integration to global brands across the US, UK, Australia, and beyond.

More information: