Somebody on your team just reported a jump in AI citations. Before anyone writes that up as a GEO win, ask what actually changed, specifically, on the site, in the weeks before the number moved. There's a decent chance the honest answer is: you finally fixed something that was broken for years, and AI had nothing to do with why it needed fixing.
Search Engine Land ran a piece on August 13 with a headline blunt enough to make the point on its own: most AI visibility gains are just technical debt repayment. The author, longtime technical SEO consultant Carolyn Shelby, walked through the mechanism plainly. Teams chase AI-specific tactics, mostly find the tactics don't move much, and then stumble into real gains while doing something far less glamorous: cleaning up redirect chains, fixing broken canonicals, untangling inherited templates nobody has touched since a platform migration two CMS generations ago. The gain gets filed under GEO optimization on the quarterly report. It was a bug fix wearing a newer label, and the label matters more than it should, because it determines what the team does next and what they think they've learned.
The pattern nobody wants to name
Here is the version of this story that has probably already happened on your team, or will soon. Someone runs an audit ahead of a GEO push. They find a chain of three redirects on category pages that has been quietly costing crawl efficiency since 2023. Nobody built it on purpose. It accreted: a URL structure change here, a migration shortcut there, a plugin update that added one more hop. Fixing it was never anyone's job specifically, because the page still resolved, the user still landed somewhere, and nothing in the existing SEO dashboard flagged it as urgent.
Then an AI crawler shows up with a five-second timeout budget and none of a browser's patience for hops. The chain that Google's infrastructure absorbed without complaint becomes the reason ClaudeBot gives up before it reaches your content. Someone fixes the redirects. Citation rate ticks up two weeks later. That improvement gets filed under GEO. It should be filed under maintenance that was already six years overdue.
Run the same story with a different symptom and the pattern repeats. A marketing team inherits a template where the actual answer to a common buyer question sits behind three collapsed accordion panels, because a designer three jobs ago thought that looked cleaner than a wall of text. Human visitors click through fine. Search engines index the collapsed text well enough because Google has spent years building tolerance for exactly that pattern. An AI retrieval crawler with no interaction model and a tight budget never gets past the collapsed state, because the text simply is not in the initial response. Somebody expands the accordion into visible text ahead of a GEO push, citations improve, and the write-up calls it an AI content restructuring win. It was a UX decision from three years ago finally getting corrected.
What counts as debt versus what counts as a tactic
The distinction matters because it changes what you should budget for and what you should expect from it. A tactic is something built specifically because AI engines exist: an llms.txt file, a schema type added purely for machine parsing, a comparison page structured around how a retrieval system extracts a verdict. Debt repayment is something you should have done regardless of AI: a working sitemap, a canonical tag that points where it says it points, a template that doesn't require six clicks to find the actual answer on the page.
| SYMPTOM | DEBT REPAYMENT (DO IT EITHER WAY) | AI-SPECIFIC TACTIC (DO IT BECAUSE AI EXISTS) |
|---|---|---|
| Redirect chains on money pages | Yes, fix regardless of AI | No new tactic needed once fixed |
| No llms.txt file | N/A | Optional; measured impact is thin, per our own [llms.txt data coverage](/blog/does-llms-txt-work-2026-data) |
| Client-side rendering hides body copy | Yes, SSR was already a UX and crawl-budget issue | Also blocks non-rendering AI crawlers entirely |
| No comparison content for top gaps | Partial, good content strategy either way | Yes, extractable verdict format is AI-specific |
| Inconsistent facts across three pages | Yes, this was always a trust problem | AI systems have no tiebreaker for conflicting claims |
Notice how much of the table lands in the debt-repayment column. That is the actual shape of Shelby's argument, and it lines up with something we have written about directly: our own agent-readable website guide opens with access and parseability precisely because those are infrastructure properties, not content decisions, and infrastructure decisions are where debt hides longest. A team that has never audited bot access, rendering, and canonicalization is not missing an AI strategy. It is missing basic technical SEO hygiene that AI simply made visible again.
Why AI exposes debt that Google tolerated
This is the part of the argument that explains why the debt was invisible for so long. Google's crawling infrastructure is enormous, patient, and has spent two and a half decades absorbing the web's mess. It follows long redirect chains. It renders JavaScript, slowly, on its own schedule. It has entity resolution systems built to reconcile inconsistent facts across a domain. None of that generosity is free, but it is Google's problem to solve, not yours, and for most sites the cost showed up as nothing worse than a slightly slower crawl.
AI retrieval crawlers operate under different constraints, and they are not shy about it. Tight timeouts and, for most of them, no JavaScript execution at all: infrastructure built to tolerate patience gets exposed the moment something arrives with none. A three-hop redirect chain is a rounding error to Googlebot and a dead end to a crawler with a five-second budget. The debt was always there. Google's patience just kept billing it somewhere you couldn't see.
The debt-versus-tactic split also explains something that otherwise looks like a contradiction: two sites can run the same AI-specific tactic, in the same category, and get wildly different results. A comparison page structured for extraction is only as good as the infrastructure underneath it. Publish a beautifully formatted comparison page on a site with an unresolved rendering problem, and the crawler never sees the formatting, because it never successfully retrieves the page's real content in the first place. This is why so many GEO retainers report disappointing early results on genuinely well-built content. The content was never the bottleneck. Nobody checked the layer beneath it.
The test: would this have helped in 2019
Shelby's piece gives technical teams a genuinely useful filter, and it is worth turning into a habit rather than a one-time read. Before crediting a win to an AI-specific tactic, ask whether the same fix would have helped your rankings, your conversion rate, or your crawl budget in 2019, back when the only retrieval system that mattered was Googlebot and the only citation that existed was a blue link. If the answer is yes, you paid down debt. Log it as maintenance, credit the team that finally got budget approval for it, and move on. If the answer is genuinely no, if the fix only makes sense because a retrieval system with no patience and no JavaScript engine now exists, you built an AI-specific tactic, and that is the smaller, harder-won category.
Paying down debt without pretending it's strategy
None of this is an argument against doing the work. Fix the redirect chains. Fix the rendering. Fix the inconsistent facts. All of it is worth doing, and Shelby's piece is not telling anyone to stop. It is asking for honesty about what the work actually is, because that honesty changes two things that matter to whoever signs off on the budget.
There is a third thing it changes too, and it is the one technical teams care about most: sequencing. If a site's citation gains are mostly waiting on debt repayment, then funding a content team to write extraction-optimized comparison pages before the infrastructure work is done is money spent on a page a crawler cannot fully reach. Shelby's ordering argument echoes what our own robots.txt and crawler-blocking data already showed: access problems are foundational and dependency-ordered, not a checklist item you can tackle in any sequence. A roadmap that funds new AI-specific content ahead of an unresolved access or rendering problem has the sequence backwards, and it will keep having the sequence backwards, no matter how good the content is, until someone runs the audit that finds the actual blocker underneath it.
First, it changes the pitch. "We have six years of unmaintained technical debt and it is now costing us AI visibility on top of everything else it was already costing us" is a harder sell than "we need an AI optimization budget," but it is the true sentence, and true sentences hold up better under a second quarter of scrutiny than exciting ones do. Second, it changes what success looks like next quarter. A team that thinks it just executed a repeatable AI tactic will look for the next tactic. A team that understands it repaid six years of debt will go looking for the rest of the debt, which is usually where the larger, more durable gains are actually sitting.
“The gain was real. The label on it was wrong. That one small distinction is worth more to your roadmap, this quarter and next, than either fact would be alone.”
Do this next: take whatever you would call your last three AI visibility wins and run each one through the 2019 test. For the ones that pass, thank your infrastructure team and stop crediting the GEO budget line for work that was really overdue maintenance. For the ones that fail, you have found the actual, smaller list of AI-specific tactics worth funding on purpose. If you have not run a bot-access and rendering audit at all, that is the first move, not the fifth, and it will very likely turn up more of your citation problem than any content project would have. Our technical SEO team runs this exact debt-versus-tactic split as the opening deliverable on new engagements, because you cannot budget honestly for either category until you know which one you are actually looking at, and most teams that ask for a GEO retainer are really asking for six years of deferred maintenance with a newer, more fundable name attached to it.
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.