Good SEO Is Good GEO. Good GEO Isn't Always Good SEO.

Nine things AI-savvy SEOs do now, specifically because AI search exists. Some are good for SEO too. Some really aren't.

At WordCamp US in August 2025, Google's Danny Sullivan put the debate to bed in five words: good SEO is good GEO. He's repeated some version of it every few months since, and he's not alone. Gary Illyes has said much the same. So has Nick Fox. In June 2026, Google formalised the line into an actual headline, publishing a piece on Think with Google called, straightforwardly, "Good SEO is good GEO."

We don't disagree with the statement as written. Good SEO probably is good GEO. The trouble is that a lot of AI-savvy SEOs are currently behaving as though the reverse is also true, and it isn't. "If good SEO, then good GEO" doesn't get you to "if good GEO, then good SEO." There's a growing list of things practitioners now do specifically because AI search exists, things nobody would bother with if ChatGPT had never shipped. Most of it happens to help traditional SEO too. Some of it doesn't, and a couple of items carry real downside.

Here's the list, and an honest verdict on each.

1. Tracking AI visibility

Rank tracking has existed for decades. Tracking whether your brand gets mentioned across a few thousand AI prompts, by industry, in something close to real time, is new. HubSpot built its own internal tool for exactly this after teams kept asking the same question: is my citation drop a strategy problem or a model update nobody told me about? Semrush has since shipped a public AI Visibility Checker doing the same job. None of this existed as a category before 2024. It's good for SEO in the sense that visibility data is always useful, but it doesn't move a single ranking on its own. It's diagnostic, not causal.

2. Running citation gap analysis

Traditional keyword gap analysis asks where a competitor ranks and you don't. Citation gap analysis asks a subtly different question: which prompts is a competitor's page getting cited for that yours isn't, even when both pages rank fine in Google. HubSpot's own research team built this into their AEO workflow specifically to catch the cases where two pages are functionally tied in Search but nowhere near tied in ChatGPT. It's a genuinely new diagnostic, and it mostly points you back toward normal content fixes, so call this one good for SEO too.

3. Building agent-ready sites

This is the one where Google is arguing with itself in public. In April 2026, Google Cloud AI's Addy Osmani published a framework called Agentic Engine Optimization, telling developers to keep pages under strict token budgets, publish llms.txt files, and serve Markdown alongside HTML so AI coding agents can parse documentation without burning through their context window. Weeks later, Google's own John Mueller called building separate Markdown pages for LLMs "a stupid idea", and Search leadership has separately warned against fragmenting content into bite-sized chunks for exactly this purpose. Two Google employees, one recommending it, one calling it daft. The honest read is that Osmani is solving for AI coding agents reading developer docs, not for Google Search rankings, and the two pieces of advice were never actually in conflict about the same job. But if you're a marketer skimming headlines, that distinction is easy to miss, and building llms.txt files for a marketing site because "Google said to" is not good SEO. It's not really good anything, yet.

4. Optimising for fan-out queries

This is the item that deserves the clearest yes on the list, not the hedge I gave it first time round. When ChatGPT or Google's AI Mode fields a complex question, it typically breaks that question into a dozen or more sub-queries before answering, a process Google's own Search Central documentation names directly as "query fan-out." A Semrush analysis run with SEO consultant Kevin Indig found ChatGPT's high-reasoning mode ran nearly five times as many background searches as its fast, minimal-reasoning mode across the same 100 prompts, 1,130 versus 245, and at the comparison stage specifically, high reasoning averaged 24 sub-queries per prompt against 5.5 for minimal reasoning. But the fix for this predates the problem by about a decade. Comprehensive topic clusters, a pillar page plus linked supporting pages that each answer one facet of a subject, are the same content model HubSpot's Anum Hussain and Cambria Davies documented back in 2015, and that model lifted rankings through internal linking alone, long before any model needed multiple queries to answer one question. Surfer SEO's analysis of 173,020 URLs found pages ranking for fan-out queries are 161% more likely to earn an AI Overview citation, and even Mike King, who has documented fan-out behaviour in more depth than most, concedes that seasoned SEOs are right to point out the mechanism existed in traditional search long before AI search gave it a name. This one is good SEO, full stop. AI search didn't invent it. It gave an old tactic a new name, a specific number attached to it, and a harder deadline.

5. Publishing more YouTube transcripts

Speaking to B&T at HubSpot's Grow 2026 conference in Sydney, Aja Frost, HubSpot's Senior Director of Global Growth, said that over half of citations for high-intent AI questions trace back to just three sources: LinkedIn, Reddit, and YouTube. Separately, Lily Ray's research into Google's AI Overviews found the same pattern for "best" queries, with Google increasingly leaning on Reddit, Forbes Advisor, and YouTube over brand-published pages. SEOs have pushed video content for years for its own sake. Publishing a transcript specifically so an LLM can lift a quote from minute six is a different motivation, even if the output looks the same. Good for SEO, mostly, since transcripts are also just accessible, indexable text. Good for the platforms hosting all that citation-worthy content, definitely.

6. Auditing AI crawler access and bot control

This one is close to home for us. A site can render perfectly for Google and still be functionally invisible to GPTBot, ClaudeBot, or PerplexityBot, either because it leans on JavaScript those crawlers don't execute, or because a robots.txt file written in 2022 blocks bots nobody had heard of yet. Osmani's AEO framework lists a robots.txt audit as step one for exactly this reason. This is unambiguously good practice, and arguably the one item on this list that's also just good technical SEO wearing a new hat, since a crawler that can't read your page couldn't index it for Google either.

7. Making facts explicit

HubSpot's own citation research found that question headers, FAQ sections, and proprietary stats are what let an answer engine actually extract a usable line from a page, rather than skip it for something plainer. There's academic backing here too: the foundational GEO paper presented at KDD 2024 found that citing statistics, expert quotes, and sources lifted visibility in AI answers by up to 40% on a position-adjusted basis. SEOs have always known that specificity beats fluff. AI search just made the penalty for vagueness immediate and measurable, rather than a slow bleed in rankings over months.

8. Boosting entity consistency across the web

Frost described this as a shift from backlinks to consensus: "are multiple independent sources all saying the same thing about your brand?" That's a different question to "who links to us," and it's a specific, additive layer that a straight SEO audit doesn't cover, since consistency in how your product is described across five unrelated sites was never a ranking factor Google scored you on directly. It only started mattering once models started synthesising an answer from all five sources at once and needing them to agree.

9. The listicle trap

Here's the one with a genuine trap in it. A study of roughly 25,000 URLs across six AI platforms found that 63% of nearly 400 million citations pointed to listicles. Listicles get cited constantly. They're structured, single-topic, and easy for a model to lift from. So publishing them looks like an obviously good AI-search move. Except Lily Ray's analysis of 100 B2B "best software" queries in Google's AI Overviews found that when a brand's own self-promotional listicle got cited, that same brand was left out of the actual recommendation 69% of the time. One real example from her research: a brand called Oasis LMS published a "best LMS" listicle naming itself alongside Kajabi, Thinkific, LearnWorlds, and Teachable. Google cited Oasis LMS's page as a source, then recommended the other four names in it, not Oasis LMS. The listicle did the work and handed the win to competitors named inside it. HubSpot's Frost makes the same point from the content-strategy side: if your AEO plan is a pile of "best X tools" posts, you're one of several names on a list everyone else is also publishing, not the answer. There's a legal wrinkle too, since ranking your own brand "best" in a self-published comparison can run into the FTC's rule against misrepresenting company-controlled content as independent reviews. Publish comparison content, but don't expect a "best X" post ranking yourself first to survive contact with an AI Overview intact.

So, is any of this good for SEO?

Start with fan-out optimisation, since it deserves the clearest answer on the list: yes, unreservedly, and it's arguably the one item here that isn't new at all, just newly urgent and freshly renamed. Crawler audits sit close behind it for the same reason. A page GPTBot can't render was never going to rank cleanly in Google either, JavaScript rendering is a shared blind spot, not an AI-only one. Explicit facts and proprietary stats are good SEO too, since specificity has always beaten filler; AI search just made the penalty for vagueness immediate rather than a slow bleed over months.

Entity consistency and citation tracking sit in the middle. Neither moves a SERP by itself, but neither does any harm. They're additive diagnostics that tell you where to point ordinary content work, not tactics with a downside.

The genuine risk sits in exactly two places: building AI-agent infrastructure for a marketing site that Google Search doesn't use ranking signals for in the first place, and publishing self-ranked listicles that get cited and then quietly handed to a competitor. Neither of those is good SEO. One is premature. The other can actively backfire, and Lily Ray's research suggests it usually does.

None of the nine are entirely new, even the ones that feel newest. SEOs have run gap analysis, pushed video, tightened crawl access, and chased consistency for twenty years, and built topic clusters for topical authority for closer to ten. AI search didn't invent most of these tactics. It raised the stakes on some by an order of magnitude, gave a couple of them a formal name and a citation-rate number to point at, and made one or two, agent-ready markup, citation tracking as its own discipline, worth doing for the first time. That's a difference in degree serious enough to change the budget line, even where it isn't a difference in kind.

One more thing worth sitting with. The message that "you don't need to do anything different for AI search" has come from Google's Search Liaison, from Google's SVP of Knowledge and Information, and, in its most recent and most CMO-facing form, from Google's VP of Search and Commerce, on Think with Google, at a moment when AI Overviews are widely reported to be decoupling impressions from clicks and Google's own ad business has an obvious interest in marketers not diverting budget toward unproven new tactics. That doesn't make the advice wrong. Sullivan's core point, that unique, well-structured content beats keyword-stuffed filler, is correct regardless of who's saying it or why. But it's worth noticing who's delivering the reassurance, and building your own AI-visibility programme on your own evidence rather than on a slogan, however many times it gets repeated.

Pay attention to the new surfaces. Just don't take the reassurance at face value.