S2: Ep 20: AI Search Reads Structure, Not Content Quality

Transcript

The following is a lightly edited transcript of the conversation. Obvious transcription errors have been corrected and the wording is otherwise preserved.

The audit that made me stop and think

Speaker: Jasmine [00:19]

Recently, you ran an AI search audit on our own website made you stop and think. Not because the finding was dramatic, it was actually a very small technical issue, but because of what it implies for any life science brand that's investing seriously in content right now. If your best work is being misread by the tools your buyers are increasingly using to build short lists, the quality of that content becomes almost irrelevant. We talk a lot with life science marketing teams who are putting real effort into thought leadership, primary research, detailed frameworks, content that takes months to develop. The assumption underneath all of that investment is that good content gets found. What your audit surfaced is that the relationship between content quality and content visibility isn't as direct as most teams assume, and that the gap between the two is growing as AI search becomes a bigger part of how buyers do their research. The cause, in our case, turned out to be five characters at the end of a URL. And I think that's either a reassuring story about an easy fix or a more uncomfortable one about how little visibility most of us have into how AI engines are actually reading our sites. So, Matt, can you please walk us through what you found?

What the SEO audit got wrong

Speaker: Matt Wilkinson [01:53]

Yeah, sure. So, I mean, this is a bit embarrassing but I ran the Claude CoWork SEO audit on strivenn.com, something I've been doing regularly as we've been evolving our branding and messaging, please do come and take a look at that, by the way. The first recommendation was to build three pillar pages: the AI Discoverability Hub, the Synthetic Customer Page, the Buyer Persona page. Problem is they already existed. All three were already in the main navigation. The AI Discoverability Hub carries primary survey data from two of our own exhibitor studies, one at LRIG Drug Discovery 2025, and one at SLAS 2026. And the synthetic customer pages, well, these are some of the deepest commercial pages on the site. And the audit recommended building them from scratch because it didn't recognise them.

Five characters at the end of a URL

Speaker: Jasmine [02:43]

And so the cause was a URL slug.

Speaker: Matt Wilkinson [02:47]

Yeah, so each pillar carried a hyphen-strivenn suffix. To a human reading the navigation, it's invisible. To an AI crawler, that tail reads as a campaign landing page rather than an authority hub. Three slugs that looked like one-off tactical assets were being treated as exactly that. There was also a small schema gap on top of that. The pillars were missing an FAQPage and BreadcrumbList tags. And three of my most important pages were effectively invisible to the AI audience that matters most right now.

AI reads a different set of signals

Speaker: Jasmine [03:20]

So the argument isn't that AI search is bad at reading content. It's that AI search reads a different set of signals that most marketing teams are optimizing for.

Speaker: Matt Wilkinson [03:32]

Yeah, absolutely. So structure, naming, markup, and relationships between pages. When those signals say tactical, the content gets classified as tactical regardless of quality. So the text is downstream of the classification. If the engine cannot place you in the right category, it won't cite you as an authority in that category. Maybe think of it like Proteus in Greek mythology. He held all the answers, but would only reveal them to those who could grasp and hold him in the right form. Your pillar pages have the answers, but if the bots cannot hold them in the right classification, those answers just never surface.

The minority-behaviour objection

Speaker: Jasmine [04:07]

So the objection we hear from life science marketing teams is that their buyers are scientists doing literature search reviews, not consumers Googling suppliers. The 18.6% of AI power user figure from our SLAS survey earlier this year, that's still less than one in five attendees. You're asking teams to restructure their site architecture from minority behavior.

Speaker: Matt Wilkinson [04:38]

It's worth putting this in context. So the 18.6% figure for a power user, that's somebody that's using AI daily. That was a single event early in 2026, but that was up from 6.6% six months earlier at AI LRIG. Now, these are not necessarily indicative of the whole population, but of course, I think they're pretty good indicators of showing that growth in AI usage is definitely on the increase. And we know from other studies that AI is being used increasingly in buying behaviors. Up to 90% of all B2B purchases now involve AI at some point. So we know that those numbers are increasing quickly. We can't tell you right now that AI search drives purchase decisions in life science at the same rate it does in other categories. I don't think anybody has that data. But what we do know is that it's increasing in importance. And we do know that as that importance increases, many of the models that we use to search, particularly those from OpenAI, are fairly lazy and they rely on the training data to answer questions. And so we really need to be found for the answers now, because we need to be in the training data for the models that come in the future, so that in the future we're found as the importance comes. So we're maybe not necessarily solving for the problems of today, but we're solving for the problems that are gonna come because that is a future that is definitely gonna happen.

Why the ROI case is future-looking

Speaker: Jasmine [06:01]

So the ROI case for structural work is partly speculative.

Speaker: Matt Wilkinson [06:06]

It's not necessarily speculative. I'd argue that it is definitely future-looking. We don't know what the ROI is going to be right now. We do know that more and more people are arriving at websites from AI searches than they did before. But we also know that when people are looking in AI searches, people are clicking through less. And so being present in that constraint set of results that AI gives us is really important. Rather than getting maybe a hundred or a thousand results in a Google search, we get three or five brands listed. So if we're not in that named set of responses, what we call citation compression, we're not actually found in AI. And so that's a really important thing.

Confidence in the audit and fixing schema at scale

Speaker: Jasmine [06:47]

So in the blog you described the fix as minimal, a few redirects and two schema tags, but you also said finding it was the hard part. If the signal was that weak, how confident are you that the audit found everything? You fixed what the tool flagged, you don't know what it missed.

Speaker: Matt Wilkinson [07:09]

If I'm using a tool to find gaps and I use Claude Cowork and an off the shelf available set of skills, most people have access to those now. And running that was much cheaper than using an SEO agency. So being able to look at that and run a test, and run it again and again and be able to check that everything is technically sound, I think is worth the time. And then making those fixes, yeah, it didn't take long. Because we use HubSpot and we're a HubSpot partner, I was able to connect up Claude to then go through and redo the schema markup on all 230 odd pages across the website agentically. So I actually managed to set it up so that the fixes that we discovered, one, I could make the changes to the URLs and make a load of changes that way, but I could also go through and update everything across the website so that the schema set those pages up for being found. Now, is there an overnight success? No. But I think that there is a real drive to get this done and get this fixed as soon as we can, so that we're setting the foundations for success in the future.

The backlog failure mode

Speaker: Jasmine [08:15]

So you separated the diagnostic from the fix. But in most life science marketing teams, the fix requires a developer and a backlog conversation. You know what happens to those conversations. They get deprioritized for six months and the team moves on. You have diagnosed the problem and done nothing about it. And obviously, I'm talking here about larger organizations where IT has a separate list of priorities from marketing.

Speaker: Matt Wilkinson [08:47]

I'd argue that more and more companies now marketing owns the website and IT supports it. You're right, it is a real failure mode. The structural fix for this is as much cultural and organizational as it is a tactical fix. What I can say is that framing this as an SEO task puts it in a low priority queue. Framing it as a revenue visibility project, you're not on the short list for deals your best content should be winning, that changes the conversation with whoever controls the backlog. Whether that's enough, I guess depends on the organization.

Reframing the fix as revenue visibility

Speaker: Jasmine [09:19]

So your recommendation for teams with slow backlogs in a sense is reframe the pitch and apply pressure with data from the citation test, but also reframe it in terms of what the value is, especially economic value.

Speaker: Matt Wilkinson [09:35]

Absolutely. The lever is available to everybody. I mean if we can show leadership that your three most important pages are invisible to the tools your buyers are already using, that's a different conversation than, we need to update our schema tags, that's something technical. Better yet, when you talk to leadership, get them to run the test. Get them to see the results themselves. That's gonna make them really sit up and look. So again, show AI, don't tell it, to quote Dr. Lisa Palmer. But yeah, we've really got to make sure that leaders see what the issue is. That's gonna be really, really key.

The brand manager's first move

Speaker: Jasmine [10:08]

Yeah, I completely agree that it's important to get the executives engaged in this so that they have firsthand knowledge of what's going on. But for the brand manager who runs the citation test and finds they're not appearing, what would you recommend is their first move?

Speaker: Matt Wilkinson [10:27]

They've got access to a tool like Claude, run the Claude Cowork SEO skill, but check the URL slugs on the most important pages. Look for anything that reads as campaign naming rather than topic authority. You know, things like branded suffixes, deep subdirectory nesting, date-based conventions. Make sure you've got clean canonical URLs that reflect topic hierarchy. Those are the sorts of things that are really, really important. We need to make sure that the AI crawlers recognise the sort of content that we're trying to promote.

The unconsidered set

Speaker: Jasmine [10:56]

So coming back to the whole Strivenn example, the concept here is the unconsidered set. And you've blogged about this before. You're not losing to a competitor on the shortlist. You're absent before the shortlist forms.

Speaker: Matt Wilkinson [11:12]

Yeah, absolutely. And I think that's really important to recognize that we have to be present to be part of that shortlist now. And so these are the table stakes for getting into that considered set. And that's really, really important.

Building in public

Speaker: Jasmine [11:26]

So your full blog is on our website, strivenn.com. And I really enjoyed this conversation, Matt. I think this is a really important topic for all of our colleagues in the life sciences, whether you're a startup or a large corporation.

Speaker: Matt Wilkinson [11:45]

Absolutely. And I hope you can all learn from my mistakes. And my somewhat embarrassment having written, I don't know however many pages on AI discoverability, to have made such a silly mistake myself. That's one of the things that I found incredibly embarrassing when I saw it. But I thought, well, what, we're trying to build in public here. And if everybody can learn from my mistakes, then fantastic.

Speaker: Jasmine [12:07]

And that goes for me as well. I've made plenty of these sorts of mistakes in product management. So I think that's the value of this podcast and the blog posts is learning from our lived experiences.

Speaker: Matt Wilkinson [12:23]

Yeah, well I hope everybody has some fun running the tests and I hope I can spare you all the embarrassment of making the mistakes that I've made. And if you find that you're making others, I'd love to learn from them so I don't repeat them as well.

Speaker: Jasmine [12:35]

Fantastic. Well, thanks again, Matt, and appreciate everybody who's been listening.

Speaker: Matt Wilkinson [12:40]

Thanks, Jasmine. Speak soon.

Q&A

How do I check whether my own pages have this problem?

Run a citation test next week. Open Claude Cowork, run the SEO skill against your three or four most important pages, and ask each engine to name the brands it would cite on your core topic. If you are absent, inspect those slugs for branded suffixes, date stamps or deep nesting. One person, an afternoon, no agency fee. The diagnostic is the cheap part, so do it before you ask anyone for budget.

I found a bad slug. Will changing the URL hurt my existing SEO?

Not if you redirect cleanly. Map each old URL to a clean canonical one that reflects topic hierarchy, set 301 redirects, and update internal links so nothing points at the dead path. In HubSpot this is a settings task, not a developer ticket. Start with the single highest-value page, confirm the redirect resolves, then roll the same pattern across the rest. Small budget, low risk, fast feedback.

How do I get this prioritised when IT controls the backlog?

Stop calling it a schema fix. Frame it as revenue visibility: your three most important pages are invisible to the tools buyers already use, so you are off shortlists your content should win. Take one screenshot of a citation test that omits your brand to your next leadership meeting and ask the decision-maker to run it themselves. Seeing the gap firsthand moves it up the queue faster than any technical request.

We have a small team and no developer. Can we do this ourselves?

Yes. The whole point of this episode is that the lever is now available to everyone. Run the audit in Claude Cowork, fix the slugs and redirects inside your CMS, and add the missing FAQPage and BreadcrumbList markup. If you are on HubSpot you can update markup across many pages agentically. Pick one pillar page next week, fix it end to end, and use it as the proof you scale from.

Our buyers are scientists doing literature searches, not Googling suppliers. Why does this matter to us?

Because you are optimising for the buyer you will have, not only the one you have today. Daily AI use among event attendees nearly tripled in six months, and lazy models lean on training data, so being cited now is how you get found later. You do not need majority behaviour to justify clean structure. Fix one authority page next week so you start entering the data future buyers will rely on.