On May 6, 2026, Microsoft’s Bing team published a technical post that quietly redefined what a search index is for. The title: Evolving role of the index: From ranking pages to supporting answers. The implication: everything you know about getting your website “indexed” may no longer be enough to appear in AI-generated answers.
Combined with Ahrefs’ recent study of 17 million citations across 7 AI platforms, the picture becomes clear. AI search isn’t a faster version of keyword ranking — it’s an entirely different system with different requirements. Here’s what changed, what the data shows, and what you need to fix.
What Microsoft’s Bing Just Published — And Why It Matters
Traditional search indexing asks one question: Which pages should a user visit? The grounding index asks a different question: What information can an AI system responsibly use to construct a response?
The Bing team’s framework introduces a new unit of value — not the page, but the groundable fact: a discrete, verifiable claim with clear provenance that a language model can cite without hallucinating or misleading.
Under this model, your website being indexed in Google or Bing is necessary but no longer sufficient. The AI layer needs to evaluate whether the specific facts on your page are:
- Fresh — a stale fact about pricing, features, or your company’s position produces a misleading response
- Corroborated — claims supported by multiple independent sources get higher confidence scores
- Non-conflicting — when two indexed sources contradict each other, the grounding system registers a conflict and may decline to cite either
- Attributable — facts with named sources, dates, and context are preferred over unsourced assertions
This matters because your content may rank on page 1 for a keyword and still be invisible in AI-generated answers if its facts can’t pass a groundability check.
The Ahrefs Data: 17 Million Citations, One Clear Signal
Ahrefs analyzed 17 million citations across ChatGPT, Perplexity, Gemini, Google AI Overviews, and Copilot. The freshness gap was significant: AI-cited content is 25.7% fresher than URLs returned in standard organic search results. AI platforms cited content averaging 1,064 days old versus 1,432 days for traditional search.
The breakdown by platform reveals how differently each AI treats recency:
- ChatGPT: cites content that is 393–458 days newer than typical Google results — the strongest recency bias of any platform studied
- Gemini: cited content averaged ~1,118 days old
- Perplexity: citations averaged ~1,166 days old
- Google AI Overviews: closest to traditional search, citing content ~1,432 days old
The practical takeaway: a post you published two years ago and never touched has minimal chance of appearing in ChatGPT’s answers, regardless of its backlink profile or keyword rankings. Google AI Overviews are more forgiving — but only if your content already ranks in the top 10. Ahrefs found that 40.58% of AI citations come directly from Google’s top 10 organic results.
Groundable vs. Rankable: The Four Factors That Now Matter
Based on Microsoft’s framework and the citation data, four content properties determine whether your page becomes a grounding source:
1. Factual Density
Content that contains specific, verifiable, sourced claims outperforms content that offers opinions and summaries. A page that says “our 2025 benchmark of 412 enterprise deployments found a 38% reduction in time-to-resolution” is groundable. A page that says “our solution dramatically improves efficiency” is not. Content with statistics, citations, and quotations achieves 30–40% higher visibility in AI responses compared to equivalent content without them.
2. Entity Consistency
AI grounding systems resolve entities — brand names, product names, people — across the web. If your website calls your product “LLMagnet” in some places and “LL Magnet” or “the plugin” in others, the model may fail to link those references to a single entity. Use consistent naming throughout every page, every meta description, and every structured data field.
3. Source Attribution
Named sources get cited. When your page references “a 2025 study by Stanford researchers” or “Gartner’s Q1 2026 report,” you create an attribution chain the AI can follow. When you say “studies show,” you give the model nothing to ground. Every statistic should have a named source and a date.
4. Conflict Avoidance
The Bing grounding framework explicitly describes conflict detection: if two indexed sources contradict each other, the system registers the conflict and may not cite either. Audit your own website for contradictions — pricing pages that disagree with feature pages, old blog posts that make claims your newer posts contradict, About pages with outdated company stats.
What Most Websites Are Getting Wrong
The gap between indexed and groundable shows up in three common patterns:
Publishing and abandoning. A blog post from 18 months ago that ranked well may have strong backlinks but contains outdated statistics. ChatGPT will favor a newer, weaker-ranked article with fresher data over your authoritative-but-stale post. The solution is not always to write new content — often it’s faster to update existing content with current data and change the publication date to reflect the revision.
Hedging instead of stating. Marketing language is designed to be safe, which makes it ungroundable. “We help businesses achieve their goals” is impossible to cite. “Our tool audits 47 AI visibility signals across ChatGPT, Perplexity, and Google AI” is citable. Every paragraph of your website should contain at least one specific, falsifiable claim.
Treating all pages as equal. Not every page needs to be a grounding source. Your contact page doesn’t need citations. But your category pages, product pages, and any page targeting an informational query should be treated as potential grounding candidates and written accordingly.
A 5-Step Groundability Audit
Here’s a practical checklist you can apply to any page on your site today:
- Check the age of your statistics. Any number older than 18 months in a fast-moving field (AI, marketing, software) is a liability. Either update it with current data or remove it. Stale statistics are the most common reason otherwise strong pages fail groundability checks.
- Count your named sources. On a 1,000-word page, aim for at least 3–5 references to named, external sources. These don’t need to be hyperlinks — just named and dated. “According to Gartner’s February 2026 report” is enough.
- Check entity consistency. Search your own website for every variation of your brand name, product name, and key terminology. Consolidate to one canonical form.
- Look for internal contradictions. Search your site for claims about your pricing, team size, customer count, or feature set and verify they’re consistent across all pages. Any contradiction you can find, the grounding index can find too.
- Add a “Last Updated” date to evergreen pages. Bing’s framework specifically calls out freshness signals. A visible last-updated date combined with genuinely updated content signals to AI crawlers that the facts are current. Updating a page’s
dateModifiedschema field and resubmitting via sitemap typically triggers re-crawling within 48–72 hours.
The Platform Asymmetry Problem
One more layer complicates GEO strategy: the 17 million citation study shows that different AI platforms pull from different source pools, and they don’t overlap as much as you’d expect. Only 11% of domains are cited by both ChatGPT and Perplexity — the two most widely used AI search tools. ChatGPT leans heavily on Wikipedia-style encyclopedic content (47.9% of its top citations), while Perplexity heavily cites Reddit (46.7%).
The grounding framework is the same across platforms — fresh, specific, attributed facts — but where those facts need to live varies. For Perplexity visibility, being discussed in forums and communities matters. For ChatGPT, having your brand referenced in long-form encyclopedic content matters more. Groundability is the floor; platform-specific distribution is the ceiling.
Where to Start
The shift from “indexed” to “groundable” is not a future threat — it’s the current state of AI search. Microsoft published the framework in May 2026 to explain why brands with strong traditional SEO are still invisible in AI answers: their content is indexed but not groundable.
Start with your highest-traffic pages. Run the 5-step audit above. Update the statistics, add named sources, fix the contradictions, and resubmit. Then measure whether those pages begin appearing in AI answers for your target queries.
If you want to see exactly where your website stands today — which pages are being cited in AI answers, which signals are missing, and what to fix first — run a free audit at ai-visibility.llmagnet.com. It checks 47 signals across ChatGPT, Perplexity, and Google AI Overviews in under 60 seconds.