Google just made the rules of AI-era SEO explicit. We removed the parts of SEO Engine that don’t match. Here’s the honest version.
On 2026-05-15, Google published its official AI Optimization Guide. It’s about as plain-language and direct as anything Google has ever released on SEO. If you’ve been paying attention to the “AI SEO” content industry over the past two years, you’ll notice the guide debunks roughly half of the popular advice in one document. SEO Engine has been honest about this stuff since day one, but a few features in our older releases were aligned with assumptions that turned out to be wrong. The week the guide dropped, we shipped 0.8.0 (deprecations) and 0.9.0 (removals) to bring everything into alignment. This article walks through what changed and why.
The headline message: the recipe is unchanged. First-hand content with a point of view, crawlable pages, write for humans. The novelty is measurement, not new optimization tricks.
The Short Version
- There is no separate “AI ranking.” AI Overviews and AI Mode use the same Google index as regular Search.
- llms.txt files don’t help. Google doesn’t fetch them. OpenAI and Anthropic don’t either.
- “AI-friendly markup” doesn’t exist as a category.
- Chunking content for AI extraction doesn’t help. Google handles nuance across multi-topic pages.
- Schema markup is for rich results, not AI search.
- What works is the SEO that’s always worked: first-hand content, write for humans, be crawlable, be indexed.
- What’s genuinely new is the measurement side: AI Overviews can eat 47% to 60% of clicks on affected queries (Pew, Ahrefs), and Google does not tell you which queries are hit.
What Google Actually Says Works
The guide’s list, paraphrased and lightly compressed:
- A unique point of view, content that stands out from existing sources
- First-hand experience over recycled tips
- Write for human readers first
- Quality images and video
- Use AI tools responsibly (don’t violate spam policies)
- Be indexed, eligible for snippets
- Be crawlable
- Semantic HTML helps (but isn’t required)
- Don’t block JavaScript-rendered content from crawlers
- Page experience: responsive, low latency, clear main vs secondary content
- Reduce duplicate content
- Manage Merchant Center / Business Profile if you have one
- Prepare for browser agents (emerging area)
The contrast example Google uses is the most useful single thing in the document. “7 Tips for First-Time Homebuyers” (commodity, recycled, won’t get cited) vs. “Why We Waived the Inspection and Saved Money” (first-hand, citable). Same topic. Completely different fate.
What Google Explicitly Debunks
Direct quote from the guide:
“You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search.”
And:
“There’s no requirement to break your content into tiny pieces for AI to better understand it.”
Roughly half the “AI SEO” content industry, ended in two paragraphs. Specifically:
- llms.txt and AI-specific markup are not used
- No need to chunk content
- No special “AI style” required, Google understands nuance and synonyms
- Inauthentic brand mentions don’t help
- Targeting every query variation is scaled-content spam
- Structured data is NOT required for AI search (it’s for rich results)
Korben (FR) has the numbers: llms.txt adoption is around 6%, and the major AI providers (OpenAI, Anthropic) don’t fetch it. So the manifest you may have spent an afternoon writing has, in practice, zero readers.
What We Changed in SEO Engine
Removed: llms.txt editor
Deprecated in 0.8.0 with a prominent warning banner, fully removed in 0.9.0. The tab, REST endpoints, and helper functions are gone. Existing files on disk are NOT auto-deleted (that would be destructive); users get a one-time admin notice offering to remove the file if SEO Engine created it.
Keeping this feature would have meant selling something that doesn’t work, which is the exact thing we promise users we won’t do. If you’ve published an llms.txt for AI visibility, you can delete it. It changes nothing.
Removed: chunkability scoring
The Content Clarity score used to weight paragraphs by whether they were sized to be “chunked” by AI extractors. Google explicitly says this isn’t needed. We replaced the 4-signal model (Chunkability, Structure, Lists, Clarity) with a 3-signal model: Structure 33pts, Lists 33pts, Clarity 34pts. The math still totals 100, the score scale is unchanged, and the AI-suggestion follow-up prompt was rewritten from “improve paragraph chunking” to “improve sentence clarity for human readers.”
Strengthened: originality and first-hand POV check
Google’s guide leads with this, so we leaned into it. We already had a 20-point penalty for low originality, but the AI prompt was focused too narrowly on detecting AI-template phrasing. The new prompt scores on Google’s exact commodity-vs-first-hand axis, with Google’s homebuyer example wording in the rubric. The AI is also told to be honest, not generous: generic content should score in the 40 to 60 range, not the 70+ “everyone gets a trophy” band.
If your post reads like a generic tip list, SEO Engine will now say so, and the Magic Fix prompt will suggest a first-hand angle instead of a polish pass.
New: JavaScript-rendering check (15-point penalty)
Google’s guide is explicit that AI can process JS-rendered content “only when it isn’t blocked,” and that “this approach is more complex.” Most users assume “if my theme renders it, Google sees it.” Often not true for SPA themes, certain page builders, or content gated by client-side fetches.
When SEO Engine scans a published post, it fetches the live URL and verifies that 2 of 3 text samples from the post body appear in the raw HTML before JavaScript runs. If they don’t, the post gets a “Main content requires JavaScript” issue and a 15-point penalty. Cached 24h per post, skipped for drafts and very short posts. Can’t be Magic-Fixed because it’s a theme or server change, not a content change.
New: AI Overview Suspects (Pro, GSC)
This is the genuinely new business reality Google’s guide creates, and the feature we’re most excited about. Ahrefs measured roughly 60% CTR loss on position-1 queries when AI Overviews appear. Pew measured 47%. Google does not tell you which queries are hit. SEO Engine has GSC integration and a position-aware CTR curve, so we can detect the signature.
Inside Quick Wins there is a new “🤖 AI Overview Suspects” bucket that surfaces queries where:
- The site ranks position ≤5 (where CTR should be strong)
- CTR is below 35% of expected for that position
- Impressions ≥ 200
- Estimated lost clicks ≥ 10
The honest framing in the UI:
“These queries rank well but click-through is unusually low, often a sign Google is summarizing the answer above your result. There is no fix that ‘beats’ AI Overviews; the play is to ensure your content is what gets cited, and to diversify traffic.”
We deliberately do NOT promise we can fix this. We surface it. That’s the honest contribution.
Honest schema copy
The Schema Markup setting description now reads:
“Automatically generate and inject JSON-LD schema markup for posts, pages, and products. Powers rich results in Google Search. Per Google’s May 2026 AI guide, schema is not required for AI search visibility, enable it for rich results, not for AI Overviews.”
We did NOT expand schema coverage in this release. Adding more schema types as an “AI feature” would have been the exact dishonesty we’re trying to avoid. Schema expansion belongs to a future rich-results roadmap, framed honestly.
What We Deliberately Did NOT Change
- AI-bot tracking (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, etc.). Unrelated to Google’s ranking position. Tracking visits is a separate, factual feature. Untouched.
- MCP support. Google’s guide doesn’t address it. Our position: MCP is for agents querying your site at inference time, not for ranking. Untouched.
- Schema expansion (FAQ, HowTo, Breadcrumb). The temptation was real, that would have been the perfect “look, an AI feature!” move. Rejected because Google explicitly says schema is not an AI signal.
- Agent-readiness scanner. Emerging area Google mentioned, but standards aren’t stable. Parked, not built.
The Headline Message
The recipe is unchanged. First-hand content with a point of view. Crawlable pages. Write for humans.
If you’ve been running a clean, honest content strategy, none of this changes anything for you. If you were chasing AI SEO tricks, you can stop. Google just told you they didn’t matter.
What is genuinely new is the measurement side. Knowing when AI Overviews are eating your clicks is the highest-leverage piece of SEO information in 2026, and nobody else gives it to you. That’s what AI Overview Suspects is for.
Things You Can Cross Off Your TODO List
- Writing an llms.txt
- “Chunking” your paragraphs for AI
- Adding “AI-friendly” markup or special meta tags
- Picking up an AEO/GEO course
- Restructuring your posts to be more “AI-extractable”
Save the time. Spend it on first-hand writing about something you actually know.
Jordy’s Take
The hardest part of running an SEO plugin in this era was not knowing what to build. Every week brought a new “you need this for AI search” claim. We were skeptical and held the line. Some of those claims sounded plausible. Chunkability did sound plausible. And yet, with Google’s guide on the table, the score is: skepticism worked.
I do not enjoy removing features. But the only thing worse than removing a feature is leaving it in to look impressive. SEO Engine is now smaller and more honest, and the AI Overview Suspects feature is the one thing I am genuinely excited about, because it solves a problem that actually exists 🙂
Questions People Ask
Should I delete my llms.txt file?
You can. It does nothing. Adoption among AI crawlers is around 6%, and Google explicitly states it isn’t used. If SEO Engine created it for you, you’ll get a one-time admin notice offering to remove it. If you wrote it manually, you can delete it from your server.
I have a course about AEO/GEO. Did I waste my money?
Probably. Per Google’s guide, those acronyms describe a category that doesn’t really exist: there is no separate AI search engine to optimize for. AI Overviews and AI Mode use the same index as regular Search. The advice that overlaps with classic SEO (write well, be crawlable, be authoritative) is still good. The advice that is “AI-specific” largely isn’t.
What about other AI search engines like Perplexity or ChatGPT search?
Each has its own crawler and citation logic, but the underlying principles are the same: high-quality, first-hand content with clear structure. SEO Engine’s AI-bot tracking shows you who’s reading you beyond Google. It’s a useful signal, but it doesn’t change how you write.
Can I “rank” in AI Overviews?
Sort of. AI Overviews cite sources that rank well in regular Search for the underlying query. So the answer to “how do I rank in AI Overviews?” is the same as the answer to “how do I rank on Google?”: be the most authoritative, useful, first-hand source for the query. There is no separate optimization.
How do I know if AI Overviews are eating my clicks?
If you have SEO Engine Pro and Google Search Console connected, the new AI Overview Suspects bucket in Quick Wins will surface affected queries automatically. The signature is: position 1-5, very low CTR (below 35% of expected for that position), at least 200 impressions, at least 10 estimated lost clicks. If you don’t have GSC connected, it’s harder to spot, you’d need to compare CTR against industry curves manually.
In Short
- Google’s May 2026 AI Optimization Guide ended the “AI SEO” speculation industry.
- There is no separate AI search to optimize for. AI Overviews use the same Google index.
- llms.txt, chunkability, AEO/GEO tricks, “AI-friendly markup”: none of it helps.
- What helps is what’s always helped: first-hand content, write for humans, be crawlable.
- What’s new is measurement: AI Overviews can eat 47-60% of your clicks invisibly. SEO Engine’s new AI Overview Suspects feature detects this.
- SEO Engine 0.9.0 removed llms.txt + chunkability, strengthened originality scoring, added a JS-rendering check, and added AI Overview Suspects detection.
Try SEO Engine on one post. Connect Search Console to see the AI Overview Suspects bucket in action. Free, no signup, ten minutes 🙂
More on the new SEO landscape: WordPress SEO in 2026: The Four Pillars and SEO Engine vs Yoast vs Rank Math, both updated for the May 2026 guide.
