Day two of SEO Week 2026, a ballroom in New York, and Wil Reynolds is telling a room full of marketers that most of what they published this year didn't matter. Not because nobody saw it. Because nobody believed it. Twenty-four hours later, on the same stage, Ross Simmonds tells the same industry the opposite kind of hard truth: sit still on this and somebody else writes your brand's memory for you. Put those two talks next to each other and you get what looks like a genuine fight over AI search content strategy — except it isn't one, once you notice they're describing two different moments in the same job.
Two stages, two warnings, one week
SEO Week runs its days as loose themes, and this year's lineup put two of the sharper voices in the industry back to back. Day two was billed as The Psychology. Day three was The Ecosystem. Reynolds, founder and CEO of Seer Interactive, took the Tuesday slot and spent it on why visibility has stopped being the finish line. Simmonds, founder and CEO of Foundation Marketing, took Wednesday and spent it on a threat most teams aren't tracking at all: the picture an AI model is quietly building of your brand, whether or not you show up to shape it.
Neither talk was a hot take built for clips. Both were the kind of argument you get from someone who has sat inside enough client accounts to know the difference between what looks like progress on a dashboard and what actually changes a buyer's mind. That's part of why the apparent contradiction between them is worth taking seriously instead of shrugging off. These aren't two hype-cycle voices talking past each other. They're two operators, describing two real and different failure modes, in a way that reads, at first, like they disagree about what to do next.
Wil Reynolds: visibility isn't the job, believability is
Reynolds' framing starts from a fact most of the room already knew and mostly ignored: zero-click results and AI Overviews have quietly detached being seen from being read. You can rank, appear, get summarized, get cited even, and still never register as a real answer in a buyer's head. His point was that marketing teams have responded to that shift by producing more of the thing that no longer works. More posts, more pages, more content in the loosest sense of the word, instead of asking what would actually move a skeptical buyer.
“The job of marketing was never to just be seen or be visible. You have to turn that visibility into something believable about your brand.”
He gave the failure mode a name that's going to stick: zombie content. Material with reach and no belief behind it. Pages that technically rank, technically get impressions, technically check a box in a content calendar, and that nobody, human or model, actually trusts. Reynolds traced it back to how a lot of content gets made in the first place, which he described as marketing whack-a-mole: teams grinding out a monthly quota of posts to keep leadership satisfied that something is happening, rather than building the smaller number of things a real buyer would actually stake a decision on.
Ross Simmonds: the model is writing your bio whether you show up or not
Simmonds picked up the next day with a different clock running. His argument isn't about whether your content is believable once someone reads it. It's about whether you get a say in what gets remembered at all. AI models aren't static reference books. They're forming and re-forming an impression of every brand in a category continuously, and that impression gets pulled into answers whether the brand contributed a single sentence to it or not. If you're not actively shaping what a model has decided about you, the gap doesn't stay empty. A competitor fills it. An aggregator fills it. Sometimes a reviewer or a forum thread you've never heard of fills it, simply because the model still has to produce some answer when it's asked.
“I think that memory is one of the most important concepts in our industry for the next 48 months. This is a pivotal moment for all of us to recognize we have an opportunity.”
The 48-month window is doing real work in that quote. Simmonds isn't describing a permanent law of the category. He's describing a land grab with a rough expiration date, the period in which how a category gets represented in model memory is still unsettled enough to influence. His diagnosis of most current thought leadership was blunt: forgettable. Competent, on-brand, forgettable. What he argued brands need instead is content built to be repeated, genuinely shareable frameworks, language distinctive enough that it has a real shot at getting picked up and baked into how a model describes the category, not just how a company describes itself.
The AI search content strategy fight that isn't actually a fight
Line those two talks up and the tension is obvious. Reynolds is telling the industry to slow down, get more selective, stop shipping things nobody believes. Simmonds is telling the industry to move now, because the window to shape model memory is measured in months, not years. Read as advice for the exact same moment, those instructions cancel each other out. Slow down and speed up isn't a strategy. It's a coin flip dressed up as guidance, and that isn't actually what either of them is telling you to do.
| REYNOLDS' WARNING | SIMMONDS' WARNING | |
|---|---|---|
| What it's a reaction to | Zero-click results and AI Overviews making visibility alone worthless | Models forming a permanent impression of your brand with or without your input |
| The core risk | Producing content nobody believes — reach without trust | Staying silent while someone else defines your category in the model's memory |
| The clock | No explicit deadline — a standing discipline | A window Simmonds puts at roughly 48 months before category representation hardens |
| The failure mode if ignored | Zombie content: technically visible, actually invisible to trust | A model-shaped bio you never wrote and can't easily edit |
Here's the read that doesn't hold up: that this is a genuine disagreement and you have to pick a side, credibility person or speed person, trust-builder or land-grabber. It's a tempting read because conference stages reward a clean binary, and because these are two people the industry already respects, which makes it feel like you're choosing a camp rather than reading two halves of one argument. But look at what each of them is actually diagnosing. Reynolds is describing what happens when you skip credibility. Simmonds is describing what happens when you skip speed. Neither of them said the other variable doesn't matter. They were standing on different days of the same conference, each naming the failure mode closest to their own client work.
The correct order: credibility, then distinctiveness
The honest version of this isn't "both matter," which is the kind of answer that sounds balanced and commits to nothing. The honest version is that they're sequenced, and the industry's actual mistake right now is running Simmonds' half of the instructions without Reynolds' half underneath it. Teams are absolutely internalizing the speed argument: publish the sharp framework, coin the memorable phrase, move before the window closes. What they're skipping is the credibility constraint that's supposed to come first — real evidence, real specificity, named sources, the unglamorous trust-building work Reynolds spent his whole talk on.
Reverse that order and you don't get a wash. You get something worse than either failure mode on its own. A distinctive, quotable, unbelievable claim doesn't just fail to land. It gets repeated. That's the part of Simmonds' argument that cuts against skipping Reynolds' constraint: memorability isn't selective about what it's attached to. A sharp, shareable framing built on a stat nobody sourced or a claim nobody checked can absolutely get picked up and folded into how a model represents your category. It's just that what gets folded in is a memorable falsehood instead of a forgettable truth. Reynolds' zombie content, it turns out, isn't always ignored. Sometimes it's exactly the kind of thing a model remembers, just not in the way anyone wanted.
That's the real cost of running the sequence backwards, and it's worse than doing nothing. Being forgotten leaves the door open. You can still show up next quarter with something better. Being remembered wrong closes it, at least for a while, because now there's an established, quotable version of your brand circulating in outputs you don't control and can't quietly retract. The fix isn't choosing Reynolds over Simmonds. It's earning the credibility first, the evidence, the specificity, the sources a model or a skeptical buyer can actually check, and only then investing in making that credible material distinctive and repeatable. Boring, verifiable, and quietly true, before clever.
A workable starting split for teams playing catch-up on both fronts, not a universal ratio
Where most AI search content strategy work skips a step
The pattern shows up the same way almost everywhere we look at it. Somebody on the team gets excited about a framework, a naming device, a three-part model, a phrase engineered to be repeatable, and it goes out before anyone has done the slower work of establishing why a reader, or a model, should believe the underlying claim. It reads well. It might even get picked up somewhere. It just isn't built on anything that survives a second look, which is exactly the gap between visible and believable that Reynolds spent his talk on.
None of this is an argument for slowing everything down indefinitely, which would just be surrendering Simmonds' half of the point. The window he's describing is real, and treating it as infinite is its own mistake. Plenty of categories are going to have their model-memory representation settle over the next couple of years whether the incumbents show up or not. The point is narrower than "be careful." The speed only pays off once there's something true and specific underneath it to be fast about. Sourced, checkable material built quickly beats unsourced material built quickly every time a model has to decide who to trust, and it beats slow, cautious, sourced material too, because slow forfeits the window Simmonds is actually right about.
This is also where a lot of GEO advice quietly goes wrong, treating generative engine optimization as a distribution problem — get the framework out, get the phrase repeated, get the citation — when it's just as much an evidentiary one. The teams doing this well right now, especially in B2B SaaS, are running something closer to a two-speed system: a steady, unglamorous stream of sourced, specific, checkable material, and a much smaller, more deliberate stream of frameworks built only on top of what's already been established as credible. That's not two competing programs. It's one program with an order of operations.
The mechanism behind it is the same one worth watching in a different frame: models don't just skip citing pages that lack credibility signals, they sometimes cite a source without actually crediting it, which is its own version of getting remembered wrong. And the instinct to chase a visibility number instead of asking whether the underlying claim would survive scrutiny is the same instinct worth pushing back on when weighing whether AI visibility tracking is a vanity metric — visibility and belief are not the same axis, on a dashboard or in a model's memory. Understanding the handful of signals that actually earn an AI citation is the credibility work Reynolds is describing. Deciding which of those citable claims deserves the shareable-framework treatment is Simmonds'.
None of this is a reason to pick a side between two people who are both right about the thing they watched happen up close. It's a reason to go look at your own pipeline and ask which stage it's actually skipping, not in theory, in the last three things you shipped. If the honest answer is that you've got plenty of frameworks and not much underneath them, you don't have a Simmonds problem. You have a Reynolds problem wearing a Simmonds costume, and a model is going to remember exactly that.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.