Improving Content Grounding in AI Search

Copilot cited GoGoChimp 1,466 times for "improving content grounding in AI search" in Bing's sample for the 90 days to 25 September 2026, and still gave nearly nine in ten of that phrase's citations to other sites.

You're in that contest whenever an AI engine answers a question in your field. It searches first and reads a few pages. Then it pins each claim to a passage that supports it, or drops the claim. Our guide to how AI engines retrieve and read pages covers the searching, so I'll stick to the pinning here.

Grounding happens a sentence at a time, and you can fix a sentence this afternoon.

What is content grounding in AI search?

Grounding is where an AI engine decides which retrieved facts it can use, and ties each claim in its answer to one of them. Bing's team puts it as one question, "what information can an AI system responsibly use to construct a response?" Bing says the unit of value in search is the page, and in grounding it's "discrete, supportable facts with clear provenance" (Bing, May 2026).

Google's Search Central guide calls retrieval-augmented generation a technique "also known as grounding", one that shows links to pages "that support the information in the response" (Google Search Central, 2026).

I read that as a job description for publishers. Give the engine each claim whole, in one dated and sourced sentence. Then make sure every page you own gives the same figure.

How is being grounded different from being retrieved, cited or mentioned?

They're four separate outcomes, and a page can win one and lose the next. No engine reports the gap between grounded and cited, so a link doesn't prove support.

Outcome	What it means	What the evidence shows
Retrieved	The AI's own search finds and reads your page	ChatGPT cited 49.98% of the URLs it retrieved across 1.4 million prompts (Linehan, 2026)
Grounded	A claim in the answer is tied to a passage on your page	Google Cloud says a response "won't be grounded" for reasons including "low source relevance or incomplete information" (Google Cloud, 2026)
Cited	A link to your page sits beside the claim	In 2023, only 74.5% of citations from four generative search engines supported their sentence (Liu, Zhang and Liang, 2023)
Mentioned	Your name appears with no link, or a link to someone else	Gemini did this in 54% of responses in a 2025 news study, against 4% or less for other assistants (EBU and BBC, 2025)

What are grounding queries?

They're the phrases an AI engine searched with, which is why they read like nobody's real question. Bing defines them as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers" (Bing, February 2026). Its help page adds that they "are not full user questions or prompts" (Bing Webmaster Tools Help).

Of Microsoft, Google, OpenAI, Anthropic and Perplexity, only Microsoft shows these phrases to every site owner. Bing Webmaster Tools is the closest you'll get to watching grounding happen on your own pages, even if it's only Microsoft's slice.

How does each AI engine pin a claim to a source?

Every AI engine that documents its citations attaches each one to a span of its answer, from part of a sentence up to a passage in Copilot Search. Where it's documented, the text taken from your page is small too, down to 150 characters for Claude.

Engine and surface	What the citation is attached to	What it points at	Source
Gemini API	A text segment, marked by start_index and end_index	A url_citation with the source URL	Google AI for Developers, 2026
Google Cloud grounding	A segment in groundingSupports, with a confidence score per supporting chunk	Web chunks from Google Search	Google Cloud, 2026
OpenAI Responses API	A url_citation annotation with start and end characters	URL and title, plus a sources list of every page consulted	OpenAI
Claude web search	A block of answer text	cited_text of up to 150 characters, plus the page_age	Anthropic
Perplexity Agent API	A numbered [n] marker in the answer	A matching search_results entry, with no character positions	Perplexity, 2026
Microsoft Copilot Search	The whole sentence or passage, linked inline	Not documented	Bing, April 2025

All but Copilot Search are developer APIs. They're the closest public evidence of how the consumer products work.

Each engine's other habits are in our notes on optimising for Perplexity and Claude and Gemini SEO.

What does Google's grounding checker count as one claim?

Google Cloud's check grounding API treats each sentence as one claim and scores its support from 0 to 1. A claim that's "only partially entailed" doesn't count as grounded (Google Cloud, 2026). In Google's worked example, a sentence names Google's founders correctly and gets the founding year wrong, so the whole sentence fails.

No Google document says AI Overviews uses this enterprise product. I'm quoting it because it shows how a grounding check reads a page.

Microsoft's Krishna Madhavan lists "Overloaded sentences" as a writing mistake, because "Packing multiple claims into one line makes it harder for AI" to parse them (Microsoft Advertising, 2025).

How often do AI engines cite pages for claims those pages don't make?

In a Washington University audit of 98,020 Google AI Overview claims from spring 2026, 11.0% weren't supported by the pages cited for them (Xu, Iqbal and Montgomery, 2026). Each audit set a different task, so don't quote a figure from one without its task and date.

What did the 2026 AI Overviews audit find?

Omission was the main failure. Haofei Xu, Umar Iqbal and Jacob Montgomery ran 55,393 trending US queries between 13 March and 21 April 2026. They split the AI Overviews into single claims. In 6.98% of claims, the cited pages didn't mention the claim at all. A cited page contradicted the claim in only 2.66%.

The authors treat 11% as an upper bound, partly because a language model did the checking. They also found source quality and claim fidelity "largely independent". So a reputable domain in the citation list proves nothing about the sentence beside it.

Are AI answers about news worse?

Yes, though that's partly because the news audits set harder tasks. In a March 2025 Tow Center test, eight AI search tools "provided incorrect answers to more than 60 percent" of 1,600 requests to name a news excerpt's source (Tow Center, 2025).

The EBU and BBC's mid-2025 study covered 22 public broadcasters in 18 countries. It found significant sourcing problems in 31% of AI assistants' news answers, and in 72% of Gemini's (EBU and BBC, 2025). In the BBC's earlier round, 13% of quotes attributed to BBC articles were "either altered from the original source or not present in the article cited" (BBC, 2025).

Why is a misattributed citation the publisher's problem?

Because the mistakes go out under your name, or under someone else's when the work was yours. In late 2024, the Tow Center caught ChatGPT search crediting a plagiarised copy of a New York Times article (Tow Center, 2024). Its 2025 test found tools citing syndicated copies on Yahoo News or AOL instead of the original publisher (Tow Center, 2025).

Losing a citation to a copy of your own article is like writing the right answer on the pub quiz sheet and watching the quizmaster give the point to the next table.

Don't expect users to complain. An ICLR 2026 study analysed more than 24,000 search conversations. People rewarded irrelevant citations almost as much as correctly attributed ones, with preference coefficients of 0.273 and 0.285 (Miroyan et al., 2025).

What does our own Bing grounding data show?

In Bing's sample for the 90 days to 25 September 2026, "improving content grounding in AI search" was the fifth-largest of our 598 grounding queries. Copilot cited us 1,466 times for it, at an 11.79% share. On four A/B testing queries in the same export, our share ran from 25.10% to 30.45%.

It was lower on every grounding query with real volume.

Grounding query (Bing's wording)	Copilot citations of GoGoChimp	GoGoChimp's citation share
improving content grounding in AI search	1,466	11.79%
strategies improving content grounding AI search	61	2.39%
improving content grounding AI search	38	10.95%
improving content grounding in AI search strategies	19	19.79%
solutions leveraging grounding for AEO comparison	9	15.79%
improving content grounding AI search strategies	4	33.33%
best A/B testing tools 2026	584	25.10%
best A/B testing platforms 2026	349	26.95%
trusted A/B testing tools for large product teams	157	29.13%
best website A/B testing tools 2026	134	30.45%

Bing Webmaster Tools AI Performance export, pulled 28 September 2026 for 28 June to 25 September 2026.

Ignore the smallest rows. A share worked out on 4 citations jumps several points with one more answer, so 33.33% doesn't mean much.

Our share was highest on the queries closest to what we sell, and 10.95% to 11.79% on the two biggest grounding rows. My guess, which Bing can't confirm, is that plenty of publishers have a passage that fits a broad explainer query.

The export hasn't got a page column, but Bing's query-to-page view does. For this phrase, all 1,466 citations went to one page, our guide to optimising for AI search engines, between 9 and 30 August 2026, with none from then to 6 October. So that 11.79% is a three-week snapshot, not a steady share.

Why does Copilot cite us for a phrase we don't rank for?

Because grounding queries are the AI's own searches, and they don't have to match what people type. On 9 October 2026, a Bing search for this phrase from a UK address didn't show gogochimp.com in the organic top ten. It's one snapshot from one place.

Which grounding tactics does the evidence support?

Relevance to the AI's own search phrase is the strongest lever, once your page is retrieved. In controlled trials, a recent date beat an old one in every model tested, and evidence-backed claims gave a smaller edge. Generic statistics, quotes and citations failed a 2026 re-test. Formatting or schema alone didn't show a reliable effect.

Tactic	Best evidence	Type	Grade
Get retrieved, and rank in the AI's own search	ChatGPT cited 58% of pages at position 0 of its search results and 14% at position 10 (Indig, 2026)	Correlational	Strong, a precondition
Match the AI's search phrase	Moving a document first in the model's context beat every rewrite in C-SEO Bench (Puerto et al., 2025). A 45-study survey calls topical relevance and context position "the most reproducible levers" (Martinez, 2026)	Controlled, plus a survey	Strong
Keep a recent date on current facts	A recent timestamp beat an old one in all six models across 252,000 trials (Vishwakarma et al., 2026)	Controlled, product reviews	Moderate
Back each claim with real evidence	Claims with evidence beat the same claims without in four of six models, in the same trials	Controlled, product reviews	Moderate, secondary
Write one whole claim per sentence	Every documented citation attaches to roughly a sentence of the answer. No study has isolated it	Vendor documentation	Indicated, untested
Add statistics, quotes or citations for their own sake	Moved citation on none of ten current model families in a September 2026 re-test (Bajemon and Rochet, 2026). Statistics rewrites lowered rankings in 19 of 24 C-SEO Bench settings	Controlled	Failed re-tests
Change formatting alone	Formatting "had no impact" in Vishwakarma et al.'s 252,000 trials	Controlled	Weak
Add schema markup	No lift in Ahrefs' matched test of 1,885 pages that added JSON-LD (Linehan, 2026)	Matched test	Null

Start with the strong rows, because sentence work can't help a page the engine never retrieves. There's more on rank in our breakdown of how ChatGPT chooses which pages to cite.

Why did the 2023 GEO levers fail the re-test?

The 2023 gains came from rewriting five fixed sources for a GPT-3.5-era model. The Princeton-led team behind the GEO benchmark reported 30-40% visibility gains for its top three methods and 41% for its best (Aggarwal et al., 2024). Our history of generative engine optimisation traces where the label came from.

In September 2026, Elisha Bajemon and André-Louis Rochet re-tested those levers as paired, length-controlled edits. None moved citation on any of ten current model families (Bajemon and Rochet, 2026). Both authors work for a company that sells a content score. Yet their paper concludes that query-blind content scores work as "a quality filter rather than a citation predictor".

I still cite every figure on this page, because readers can check it.

Where do Google and Microsoft disagree?

They disagree on whether to shape text for AI at all. Google says "You don't need to write in a specific way just for generative AI search" and "There's no requirement to break your content into tiny pieces" (Google Search Central, 2026). It also tells site owners to ignore llms.txt, which our llms.txt explainer goes into.

Microsoft asks for "Self-contained phrasing", sentences "that make sense even when pulled out of context", and lists "Unanchored claims" as a mistake. It also warns that "AI systems may not render hidden content" in tabs (Microsoft Advertising, 2025).

A complete, sourced sentence written for a person satisfies both. For placement, see our guide to where the answer should sit on the page.

How do you improve content grounding, step by step?

Work on one grounding query at a time, in the order the evidence supports. Make the page the best match for the AI's phrase first. Then fix the sentences that carry its claims, and remove every competing version of those facts.

In Bing Webmaster Tools' AI Performance report, sort your grounding queries by citations and pick the biggest one where your share is low.
Select it and open the pages Bing maps to it, a view Microsoft added in March 2026 (Madhavan and Merchant, 2026).
Make sure the cited page answers the AI's phrase, with a heading close to its wording and the answer straight under it. Our long-tail AEO playbook shows how to cover follow-up phrases on one page.
Rewrite each sentence that carries a claim so it holds the figure, what it measures, the date and the source.
When a fact changes, update the sentence and the visible date together.
Search your site for older versions of each figure, and correct or retire them.
Ask anyone who republishes your article for a canonical tag pointing at your original, or give them an excerpt instead.
Keep facts in the page's HTML text, not in a PDF, an image or a closed tab.
After 30 days, compare the query's citation share with the previous 30.

What does a groundable sentence look like?

It holds one claim and everything a checker needs to confirm it.

Before: "Our share on grounding queries is about 12%, way below what we get on testing tools."

After: "In Bing's sample for the 90 days to 25 September 2026, Copilot gave GoGoChimp 11.79% of the citations it showed for 'improving content grounding in AI search'."

The first packs in two claims with no date or source, and compares against a figure the reader can't see. Anyone with Bing's export could check the rewrite on its own.

What do stale dates and conflicting figures cost you?

They give the engine a reason to drop you, or to say something that isn't true in your name. Bing says "a stale fact produces a misleading response". Faced with contradictory sources, a system may "confidently assert the wrong thing". It may also abstain, which Bing calls valid when support is "missing, stale, or conflicting" (Bing, May 2026).

The BBC caught Copilot using a 2022 live page as its only source for a current question (BBC, 2025). In Sprinklr's 252,000 controlled trials, internal contradictions were a secondary factor in which source got cited first (Vishwakarma et al., 2026).

Age alone isn't the problem. Ahrefs found Google's top three AI Overview citations averaged 1,432 days old (Law, 2025). Change the date when the facts change, and only then.

I've spent part of this month sweeping superseded figures out of our own live posts. One outdated number kept turning up like a bad penny, post after post.

How do syndicated copies and duplicates cost you grounding?

They're competing with your original for one slot. Bing says near-duplicates reduce "the chances that your preferred page will be chosen as a grounding source". The model may also pick "a version that is outdated" (Bing, December 2025). Its fix is step 7, a canonical tag on the partner's copy or an excerpt instead of the full article.

How do you find ungroundable sentences with Rubric?

Rubric, GoGoChimp's AI citability auditor, crawls your site and looks for unsourced statistics, missing author names, pages with fewer than two outside sources, sections that don't stand alone and pages that have gone stale. Rubric predicts citability, it doesn't count citations, so use it to find pages to fix and Bing to see what changed.

Run a free Rubric audit on up to 25 pages, with no sign-up. It scores 44 checks, grouped into Known, Findable and Trusted (will an engine quote you). On the Overview, look at the Content citability signals card and the Decay risk card, which groups pages by age.

What did Rubric find on our own site?

It found pages to fix on a site that scored 89 overall.

On our own 182-page crawl of 5 October 2026, Rubric's Trusted pillar scored 88. Content citability signals flagged 3 pages with unresolved passages, 103 with unbacked superlatives and 10 with stale or mismatched schema values.

Across the 244 sites scored in Rubric's benchmark re-crawl of 7 October 2026, the median site scored 66, and 189 of them (77%) sat below 70, the score we treat as quotable. Trusted was the lowest of the three pillars, with a median of 49. At 138 sites (57%), more than half the applicable pages had fewer than two external source links.

Use the score as a filter for pages worth fixing, which is how the September 2026 re-test says to use any content score.

How do you confirm a grounding fix in Bing Webmaster Tools?

Compare the same grounding query's citation share across two periods. Bing Webmaster Tools added citation share and a Compare view in June 2026. It defines share as "the percentage of citations attributed to your site" out of every citation shown for that query, with competitor domains hidden (Bing, June 2026). Don't read the count without the share, and treat any change as a correlation.

The report covers "Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations". Bing calls its data "a sample of overall citation activity" (Bing, February 2026). The partners aren't named, so don't assume ChatGPT is among them. Our walkthrough of Copilot SEO goes through the rest of the report.

In Bing's sample, our citation share for "best Shopify CRO agencies UK" was 62.75% in the 90 days to 1 July 2026 and 40.00% in the 5 October 2026 export. Our citations rose from 32 to 112 over that stretch, and other sites' citations rose faster.

What can Search Console and prompt trackers tell you?

Search Console only counts impressions. Its generative AI report has been live for every site worldwide since 31 August 2026. It splits AI Overviews, AI Mode and Discover impressions by page, country and device, but it doesn't show queries, citations or separate clicks (Google Search Console Help).

Prompt trackers fill the gap for ChatGPT, Claude and Perplexity, which don't offer a citation report open to every site. The answers they log keep moving. SparkToro found less than a one-in-100 chance that ChatGPT or Google's AI returns the same brand list twice in 100 runs (Fishkin, 2026). Ahrefs gives an AI Overview "a 70% chance of changing from one observation to the next" (Linehan, 2025).

We compare the trackers in our roundup of AI SEO monitoring tools. For sampling, read our guide to tracking AI visibility over repeated runs.

How do you check whether the cited sentence is supported?

By hand, because there's no first-party tool that checks it. Ask the engines the questions your grounding queries point to, several times each. Compare every sentence credited to your page with what the page says. Log omissions and contradictions, the two biggest failure types in the 2026 AI Overviews audit, plus any credit that went to a copy.

Run it on your biggest low-share query before you rewrite the page and again 30 days after. To start on the page side, run a Rubric audit on the pages Bing maps to that query. The rest of our AI search guides are in the AI SEO library.

Questions about content grounding

Is grounding the same as RAG?

Not quite, though Google uses the two terms almost interchangeably. Retrieval-augmented generation is the whole technique. Grounding is the step that ties each claim to evidence. OpenAI's API lists consulted sources separately from cited ones (OpenAI), so a page can be retrieved and read without grounding anything.

Does grounding stop AI hallucinations?

No, though it reduces them. In Google DeepMind's December 2025 FACTS benchmark suite, no model scored above 70% overall. The best grounding score was 74.2, on tasks where the source material is supplied (Cheng et al., 2025). OpenAI says ChatGPT's search results and citations can be "incomplete, outdated, or incorrect" (OpenAI Help Center).

Do all AI engines search the web before answering?

No. Gemini's API first decides whether a Google Search "can improve the answer" (Google AI for Developers, 2026). In logged search-chat sessions from 2025, 34% of Gemini's responses and 24% of GPT-4o's were written without fetching any online content (Strauss et al., 2025). An answer written without a search hasn't got a web page to ground on.

Why does my high-ranking page not appear in AI answers?

Because the AI runs its own searches and picks its own sources. In the 2026 Washington University audit, nearly 30% of domains cited in AI Overviews didn't appear on the first page of organic results at all (Xu, Iqbal and Montgomery, 2026). Rank for the AI's grounding query as well as your customer's search.

Does schema markup improve grounding?

No study we've seen shows it does. In Ahrefs' matched test of 1,885 pages that added JSON-LD, AI Overview citations fell 4.6%. ChatGPT's 2.2% rise was indistinguishable from zero (Linehan, 2026). Google says you don't need special schema for its AI features (Google Search Central, 2025). Keep schema accurate and matched to the visible text.

Does blocking AI crawlers or opting out of AI features stop my content being used for grounding?

Only partly. Google's generative AI control, rolled out worldwide on 31 August 2026, makes opted-out content ineligible "to be used as an input" for AI responses (Google Search Console Help). OpenAI says a page that blocks OAI-SearchBot may still appear as a link and title (OpenAI Help Center). OpenAI's crawlers are covered in our guide to getting cited by ChatGPT.

References

Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative Engine Optimization. KDD 2024. arXiv:2311.09735. https://arxiv.org/abs/2311.09735
Anthropic. (n.d.). Web search tool (Claude Platform documentation, read 9 October 2026). https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool
Bajemon, E., & Rochet, A.-L. (2026, 7 September). Scoring Without the Engine: Validating a Deterministic, Manipulation-Resistant Content Score for Generative Engines, End to End. arXiv:2609.07559 (preprint). https://arxiv.org/abs/2609.07559
BBC. (2025, February). Representation of BBC News content in AI Assistants (fieldwork December 2024). https://www.bbc.co.uk/aboutthebbc/documents/bbc-research-into-ai-assistants.pdf
The Bing Team. (2025, 4 April). Introducing Copilot Search in Bing. Bing Search Blog. https://blogs.bing.com/search/2025/4/Introducing-Copilot-Search-in-Bing/
Canel, F., & Madhavan, K. (2025, 19 December). Does Duplicate Content Hurt SEO and AI Search Visibility? Bing Webmaster Blog. https://blogs.bing.com/webmaster/2025/12/Does-Duplicate-Content-Hurt-SEO-and-AI-Search-Visibility/
Cheng, A., Jacovi, A., Globerson, A., et al. (2025, 11 December). The FACTS Leaderboard: A Comprehensive Benchmark for Large Language Model Factuality. Google DeepMind. https://storage.googleapis.com/deepmind-media/FACTS/FACTS_benchmark_suite_paper.pdf
Fishkin, R. (2026, 28 January, updated 28 June). SparkToro research on how inconsistently AI tools recommend brands and products, and what that means for tracking AI visibility. https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
Fletcher, J., & Verckist, D. (2025, October). News Integrity in AI Assistants. European Broadcasting Union and BBC. https://www.ebu.ch/files/live/sites/ebu/files/Publications/MIS/open/EBU-MIS-BBC_News_Integrity_in_AI_Assistants_Report_2025.pdf
GoGoChimp. (2026, 28 September). Bing Webmaster Tools AI Performance exports for gogochimp.com (AI search queries report and overview), covering 28 June to 25 September 2026. Also the early-July and 5 October 2026 exports for "best Shopify CRO agencies UK". First-party, unpublished.
GoGoChimp. (2026, 9 October). Bing results page snapshot for "improving content grounding in AI search" from a UK address. First-party, unpublished.
GoGoChimp. (2026, 5 October). Rubric hosted crawl of gogochimp.com, 182 pages. First-party, unpublished.
GoGoChimp. (2026, 7 October). Rubric benchmark re-crawl, 244 of 299 sites scored on scoring version 2026-10-01. First-party, summarised at /research/state-of-ai-citability-2026.
Google AI for Developers. (2026, updated 23 September). Grounding with Google Search (Gemini API documentation). https://ai.google.dev/gemini-api/docs/google-search
Google Cloud. (2026, updated 7 October). Check grounding with RAG (Agent Search documentation). https://docs.cloud.google.com/generative-ai-app-builder/docs/check-grounding
Google Cloud. (2026, updated 7 October). Grounding with Google Search (Gemini Enterprise Agent Platform documentation). https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/grounding/grounding-with-google-search
Google Search Central. (2025, updated 10 December). AI features and your website. https://developers.google.com/search/docs/appearance/ai-features
Google Search Central. (2026, 15 May, updated 10 July). Optimizing your website for generative AI features on Google Search. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
Google Search Console Help. (2026). Generative AI performance report (Search). https://support.google.com/webmasters/answer/16984139
Google Search Console Help. (2026). Search generative AI control. https://support.google.com/webmasters/answer/16908024
Indig, K. (2026, 14 April). Shorter, Focused Content Wins In ChatGPT (AirOps data). Search Engine Journal. https://www.searchenginejournal.com/shorter-focused-content-wins-in-chatgpt/571857/
Jaźwińska, K., & Chandrasekar, A. (2024, 27 November). How ChatGPT Search (Mis)represents Publisher Content. Columbia Journalism Review, Tow Center for Digital Journalism. https://www.cjr.org/tow_center/how-chatgpt-misrepresents-publisher-content.php
Jaźwińska, K., & Chandrasekar, A. (2025, 6 March). AI Search Has A Citation Problem. Columbia Journalism Review, Tow Center for Digital Journalism. https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php
Law, R. (2025, 28 July). New Study: AI Assistants Prefer to Cite "Fresher" Content (17 Million Citations Analyzed). Ahrefs. https://ahrefs.com/blog/do-ai-assistants-prefer-to-cite-fresh-content/
Linehan, L. (2025, 11 November, updated 12 August 2026). AI Overviews Change Every 2 Days (But Never Change Their Mind). Ahrefs. https://ahrefs.com/blog/ai-overview-change/
Linehan, L. (2026, 15 April). Why ChatGPT Cites One Page Over Another (Study of 1.4M Prompts). Ahrefs. https://ahrefs.com/blog/why-chatgpt-cites-pages/
Linehan, L. (2026, 11 May). We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved. Ahrefs. https://ahrefs.com/blog/schema-ai-citations/
Liu, N. F., Zhang, T., & Liang, P. (2023). Evaluating Verifiability in Generative Search Engines. Findings of EMNLP 2023. arXiv:2304.09848. https://arxiv.org/abs/2304.09848
Madhavan, K. (2025, 8 October). Optimizing Your Content for Inclusion in AI Search Answers. Microsoft Advertising Blog. https://about.ads.microsoft.com/en/blog/post/october-2025/optimizing-your-content-for-inclusion-in-ai-search-answers
Madhavan, K., & Merchant, M. (2026, 23 March). The AI Performance dashboard: Your view into where your brand appears across the AI web. Microsoft Advertising Blog. https://about.ads.microsoft.com/en/blog/post/march-2026/the-ai-performance-dashboard-your-view-into-where-your-brand-appears-across-the-ai-web
Madhavan, K., Merchant, M., Canel, F., & Nigam, S. (2026, 10 February). Introducing AI Performance in Bing Webmaster Tools Public Preview. Bing Webmaster Blog. https://blogs.bing.com/webmaster/2026/2/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview/
Madhavan, K., Merchant, M., Nigam, S., & Shah, T. (2026, 16 June). New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare. Bing Search Blog. https://blogs.bing.com/search/2026/6/New-AI-Visibility-Insights-in-Bing-Webmaster-Tools-Intents-Topics-Citation-Share-Compare/
Madhavan, K., Risvik, K., & Merchant, M. (2026, 6 May). Evolving role of the index: From ranking pages to supporting answers. Bing Search Blog. https://blogs.bing.com/search/2026/5/Evolving-role-of-the-index-From-ranking-pages-to-supporting-answers/
Martinez, O. (2026, 15 July). Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026). arXiv:2607.14035 (preprint). https://arxiv.org/abs/2607.14035
Microsoft Bing. (n.d.). AI Performance in Bing Webmaster Tools (help page). https://www.bing.com/webmasters/help/ai-performance-9f8e7d6c
Miroyan, M., Wu, T.-H., King, L., et al. (2025). Search Arena: Analyzing Search-Augmented LLMs. arXiv:2506.05334, ICLR 2026. https://arxiv.org/abs/2506.05334
OpenAI. (n.d.). Web search (OpenAI API documentation, read 9 October 2026). https://developers.openai.com/api/docs/guides/tools-web-search
OpenAI Help Center. (2026, updated 21 August). Searching the web with ChatGPT. https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt
OpenAI Help Center. (2026, updated 29 August). Publishers and Developers FAQ. https://help.openai.com/en/articles/12627856-publishers-and-developers-faq
Perplexity. (2026). How to migrate from Sonar (Perplexity documentation). https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/how-to
Puerto, H., Gubri, M., Green, T., Oh, S. J., & Yun, S. (2025). C-SEO Bench: Does Conversational SEO Work? NeurIPS 2025 Datasets and Benchmarks. arXiv:2506.11097. https://arxiv.org/abs/2506.11097
Strauss, I., Yang, J., O'Reilly, T., Rosenblat, S., & Moure, I. (2025). The Attribution Crisis in LLM Search Results. arXiv:2508.00838. https://arxiv.org/abs/2508.00838
Vishwakarma, R., Kumar, S., & Jamidar, R. (2026). What Gets Cited: Competitive GEO in AI Answer Engines. SIGIR 2026. arXiv:2605.25517. https://arxiv.org/abs/2605.25517
Xu, H., Iqbal, U., & Montgomery, J. M. (2026). Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact. arXiv:2605.14021 (v2, 1 October 2026), accepted to ACM IMC 2026. https://arxiv.org/abs/2605.14021

Improving Content Grounding in AI Search: How to Be the Sentence an AI Answer Rests On

What is content grounding in AI search?