AI Search Ranking Factors: Does Your SEO History Even Count?

Ask three people whether your old SEO history still counts toward AI search ranking factors and you'll get three confident, contradictory answers. One will point at your backlink profile and domain age like nothing changed. Another will tell you AI search is a clean slate, and your fifteen years of authority mean nothing to a chatbot. A third will say the whole question is malformed, because some AI systems don't keep a memory of your site at all. The uncomfortable truth is that all three are right, just about different platforms, and most teams are answering the wrong version of the question.

That's the finding buried in Duane Forrester's July 12, 2026 piece, "Do the Answer Engines Keep Your Fingerprint, or Do They Start Fresh Every Time?" Forrester spent years building search products at Bing and Yext, and he asked a specific, practical question: do the classic signals your team spent years building, backlinks, click depth, Core Web Vitals, authorship, domain age, schema coverage, still function as a domain-level reputation once search moves from ranked lists to generated answers? His answer isn't a single verdict. It's a platform-by-platform breakdown, and it maps almost exactly onto the three camps currently arguing past each other online.

The debate over AI search ranking factors: three camps, one platform-by-platform answer

Strip away the platform-specific nuance and the public argument over AI search ranking factors collapses into three positions, each with real evidence behind it and each incomplete on its own.

Most of the discourse treats these three as competing theories where one has to be wrong. Forrester's platform-by-platform answer, read alongside Kevin Indig's and Mike King's work, shows they're not competing at all. They're describing three different technical realities that happen to coexist inside the same phrase, "AI search."

Camp 1: Google and Bing prove AI search ranking factors carry over your SEO history

On Google, Forrester's case is architectural, not speculative. AI Mode is rooted in Google's existing core ranking systems, the same index, the same signals pipeline that has always powered classic search. There isn't a separate, independent scoring layer that ignores everything Google already knows about your domain. If Google's systems already track your backlink profile, your click depth, your Core Web Vitals, your authorship signals, your domain age, and your schema coverage, and AI Mode draws from that same infrastructure, then the fingerprint doesn't need to be rebuilt. It was never erased.

Bing tells a similar story, through a different mechanism. Records persist there too, via integration with tools like IndexNow, Microsoft's real-time indexing protocol that keeps Bing's understanding of a domain current as pages change. A domain with an established technical and authority footprint on Bing carries that footprint into Bing's AI surfaces for the same structural reason Google's does: the AI layer isn't a clean-room rebuild, it's an extension of infrastructure that already has your history in it.

Camp 2: Kevin Indig's data says AI search is a clean slate

If historical SEO authority carried over cleanly everywhere, incumbents would already own most AI-search categories, the same way they own most Google page-one results. They don't. Kevin Indig's topical authority research ran 1,094 categories through ChatGPT and found only 15.2% have a stable, consistent citation owner. The other 89.3% of estimated demand has no settled leader at all.

That gap is the strongest available evidence for the clean-slate camp. A category with a dominant, decades-old brand and a mountain of backlinks should be an easy win for the incumbent if old-school authority translated one-to-one into AI-search ownership. Instead, most categories are still unsettled, being answered by whichever brand, directory, or comparison page happens to give the model the cleanest answer that week. If your read on AI search ranking factors starts and ends with "my domain authority will carry me," Indig's number is the fact that should worry you: 84.8% of categories haven't produced a durable leader yet, which means old authority isn't automatically doing the job people assume it's doing.

This is also where Forrester's ChatGPT finding lines up almost exactly with Indig's data instead of contradicting it. Forrester calls the ChatGPT question "genuinely opaque." ChatGPT's web-grounding draws partly on Bing's index, so some indirect persistence may flow through that pipeline. But whether the model itself has independently encoded a domain-level trust score into its own weights, something functioning like the fingerprint Camp 1 describes on Google and Bing, remains unproven. In Forrester's own framing, plausible is not demonstrated. Indig's 89.3% unowned figure is what that opacity looks like in practice: if a durable per-domain reputation were quietly baked into ChatGPT the way it is into Google's ranking pipeline, you'd expect established authorities to already be winning most categories. They aren't.

Camp 3: agentic retrieval doesn't run on a reputation model at all

The third camp doesn't argue your fingerprint is erased. It argues the question is the wrong shape for certain systems, because those systems were never built to store one. This is the implicit claim behind Mike King's agentic RAG framing, which we covered in detail in our piece on the technical differences between AI crawlers and AI agents: live agentic retrieval plans, retrieves, reads, and re-retrieves multiple times within a single query, rather than leaning on a fixed, pre-computed trust score decided once and cached.

Think about what that means for the fingerprint question. A system built around a stored domain-level reputation checks that score once, then trusts it until the next crawl updates it. A system built around live agentic re-evaluation doesn't work that way. It goes back to your page, mid-conversation, and reads it again, deciding fresh each time whether this specific paragraph answers this specific question well enough to cite. There's no persistent record being consulted, because the architecture doesn't store one to consult. For those systems, asking "does my SEO history carry over" is a bit like asking whether your credit score matters to a cashier who re-checks your wallet every single time you pay, in cash, regardless of what happened yesterday.

“Plausible is not demonstrated.”

Put plainly: Camp 3 isn't claiming your reputation was reset to zero. It's claiming that for systems doing live, multi-pass agentic retrieval, there may never have been a persistent domain-level scoreboard in the first place. That's a different, and in some ways more freeing, situation than the one Camp 2 describes. A clean slate implies you're behind. A live re-evaluation model implies the page in front of the model right now is what matters most, which is a fight you can win today regardless of what your domain looked like five years ago.

The verdict: audit your AI search ranking factors platform by platform

This is also the honest answer to the broader question a lot of clients ask us directly: does SEO still matter? Yes, on the platforms where the AI layer inherits the ranking infrastructure you've already invested in. Less certainly, and possibly not at all in the way you'd hope, on the platforms that don't. Pretending there's a single answer is how teams end up complacent about ChatGPT because their Google AI Mode numbers look fine, or panic about Google because a ChatGPT citation test came back cold. Neither reaction is calibrated to what Forrester actually found.

The practical version of this verdict is a per-platform audit, not a single AI-visibility scorecard. On Google and Bing, keep doing the classic-SEO work, backlinks, Core Web Vitals, schema coverage, authorship, because that work is provably still being read into the systems generating AI answers. On ChatGPT, don't assume your accumulated authority is doing anything until you've tested it directly: run the same buyer questions repeatedly and see whether a consistent source wins, the way Kevin Indig's category-ownership study tested this at scale. On any platform running live agentic retrieval, shift the investment toward making each individual page a complete, self-contained, re-retrievable answer, since that's the unit the model is actually re-evaluating, not your domain's résumé.

None of this is measurable from a single dashboard number, which is exactly why it gets flattened into one bucket so often. Real answers require attributing pipeline to organic and AI-cited traffic separately by source, and reporting on AI search visibility platform by platform instead of blending it into one line. That's the same instrumentation gap we closed for our work with Arnica, where the client needed to know not just whether they were getting cited, but which platforms that citation was and wasn't coming from before they'd commit further budget to generative engine optimization.

Here's the specific next step. Pull your last three months of AI-citation tracking and split it by platform instead of reading it as one number. For Google AI Mode and Bing/Copilot, cross-reference citation presence against your existing backlink and Core Web Vitals data; if the correlation is strong, that confirms the fingerprint is doing what Forrester says it's doing, and the move is to keep funding classic technical and authority work. For ChatGPT, run the same set of ten buyer questions weekly for a month and log whether the cited source changes; if it does, you're inside Indig's unowned 84.8%, and no amount of legacy authority is going to fix that on its own; you need fresh, citable content built for the category, which is exactly the gap generative engine optimization services and our reporting and analytics service are built to close together, one platform at a time instead of one blended average.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.