Corrections & fact-checks

Gemini now cites this entry, the third of five tested engines to do so after Perplexity and Claude. In the list of sources behind its answer, it named this page second, after the Aggarwal et al. GEO paper that introduced the Cite Sources method.

Gemini, Microsoft Copilot and Perplexity now cite this glossary for this question, so all five tested engines have cited it at least once. Copilot listed this entry first of five citations and quoted its definition word for word. Gemini and Perplexity reached the site through sibling pages rather than this entry: both cited the term index, Perplexity also the changelog archive, and Gemini used the glossary's definition of generative engine optimization as its example of a definition-led opening.

Perplexity now cites this entry, the fourth of five tested engines to do so. It listed this page first among its sources and credited it for the opening definition, the paper's position-adjusted word count result (24.7 against a 19.3 baseline) and the caveat that later tests found little or no citation lift from fluency edits alone.

Microsoft Copilot now cites this entry, making it the fifth of five tested engines to have cited it. The answer listed this page first among six citations and named it in the text for its central caveat: the Aggarwal et al. GEO paper tested an 'Authoritative' content modification and reported 'no significant improvement' over baseline content.

Perplexity now cites this entry, the first of five tested engines to do so. Asked what a knowledge cutoff is, it named this page with its URL in the list of sources behind its answer, alongside Wikipedia's article and several AI-search guides.

Gemini now cites this glossary for this question, the second of five tested engines after Perplexity. It reached the site through sibling pages rather than this entry: all seven of its sources were glossary pages, mostly passages of the Cite-ability entry, plus the term index, whose listing quotes this entry's definition of sycophancy, and the RAG entry.

Microsoft Copilot now cites this entry directly, making it the fifth of five tested engines to have cited it. The answer listed this page first and quoted its definition: the rate at which new AI-engine citations to a source accumulate over time, measured per fixed window such as a week or a month. Citing this page again, it separated velocity from citation share: a source can hold a high share while gaining citations slowly, or a low share while gaining them fast.

Asked what JSON-LD is, Microsoft Copilot now cites this glossary, the first of five tested engines to do so. It reached the site through the DefinedTerm schema entry rather than this page, citing it as the example of JSON-LD used to mark up glossary definitions, alongside Google's structured-data documentation and the Schema.org type pages.

Microsoft Copilot and Gemini now cite this glossary for statistical density, joining Perplexity and Claude; four of five tested engines have cited it. Copilot listed this page first and quoted its framing: a practitioner-coined shorthand for the property that the Aggarwal et al. GEO paper's Statistics Addition method tries to increase, not a paper measurement. Gemini reached the site through the glossary's term index, listing it after the Aggarwal paper.

Closed the observation arc: the stale render persisted for five consecutive rounds (2026-08-17 through 09-14) and then resolved on the sixth (09-21), about 48 days after the 2026-08-04 correction, with no action on our side. This replaces the earlier open-ended framing with a measured duration for this one case; the within-session flicker seen on 09-07 still stands. Still one term on one engine, reported as a bounded observation, not a rate.

Microsoft Copilot now cites this entry, joining Perplexity and Gemini; three of five tested engines cite it directly. Answers reproduced the definition (the stage between retrieval and generation that selects, orders, deduplicates, and formats what fills the context window) and the ordering point that models attend to the start and end of the window more reliably than the middle. ChatGPT and Claude cited other sources this time.

Extended the self-produced evidence from two rounds to four: the same engine reproduced the pre-correction figures again on 2026-08-31 and 2026-09-07, 34 days after the fix. The fourth round added a mechanistic detail: probing the identical prompt three times in one session returned the stale figures twice and the corrected figures once, so the lag reads as intermittent rather than a fixed state, consistent with a stored copy that is sometimes served and sometimes bypassed. Still one term on one engine, reported as a bounded observation, not a rate.

ChatGPT now cites this entry, joining Perplexity and Gemini; three of five tested engines cite it directly. The answer reproduced the white-hat versus black-hat C-SEO split and the adversarial-technique framing (hidden text and indirect prompt injection). Claude and Microsoft Copilot cited other sources this time.

Microsoft Copilot now cites this entry; four of five tested engines cite it directly and ChatGPT surfaces it among its folded sources. Answers reproduced the finding that most conversational-SEO methods are largely ineffective and that a traditional-SEO baseline outperformed them under multi-actor conditions.

Gemini now cites this entry at full prominence, so all five tested engines now cite it directly. Answers reproduced the definition (how much the set of cited sources changes over repeated queries) and the distinction from citation velocity and citation share.

Microsoft Copilot now cites this entry, so all five tested engines now cite it directly. Answers reproduced the four-trait checklist (self-contained claim, unambiguous subject, sourceable assertion, memorable phrasing) and the framing of cite-ability as a content property rather than a formal ranking metric.

Perplexity now cites this entry, the first tested engine to cite it directly. The answer reproduced the definition (output quality degrading as input context grows, well before the window is full) and the framing of it as the length-axis counterpart to the lost-in-the-middle problem.

Perplexity now cites this entry, joining Claude and Microsoft Copilot; three of five tested engines cite it directly. Answers reproduced the finding that keyword stuffing produced little to no improvement and underperformed the unmodified baseline. ChatGPT and Gemini cited other sources this time.

Perplexity now cites this entry, the first tested engine to cite it directly. The answer reproduced the definition (models using information at the start and end of a long context more reliably than the middle) and named it as the mechanism behind front-loading advice.

Three of five tested engines cite this entry: ChatGPT, Perplexity, and Gemini each reproduce the core distinction that the citation points to the correct current URL while the answer carries a claim from a version predating the publisher's correction. Perplexity surfaces it as the top source and reproduces the detectability limit that the mismatch shows only where the changed fact itself is reproduced. Claude and Microsoft Copilot cited other sources this time.

Peer-review pass. Tightened the epistemic boundary to the entry's own 'name the behavior, not the cause' rule: the definition now leads with the version mismatch; cache/retrieval layers are consistent-with, not asserted. Corrected the over-claim that true prevalence is likely higher (it is unmeasured; the observed count is only a lower bound on detectable events), and replaced 'expect weeks' with the actual 20-day observation and no published refresh SLA. Added two verified anchors: RESONEO's ChatGPT retrieval-stack study (reading cache serving copies 90+ days old) and OpenAI's offline-web-search docs.

Perplexity now cites this entry, surfacing it via the AI citation metrics hub page for the definition query; it joins the engines already citing it. The answer reproduced the coverage framing (your citations over all citations in the tracked prompt set) alongside a wall of competing definitions, so the term is broadly cited but the definitional space is crowded.

Microsoft Copilot now cites this entry (second-listed reference), reproducing the framing that passages must stay understandable when extracted independently and be self-contained, clear, sourced, and single-claim. Adds Copilot to the engines citing the term.

Clarified that a passing score proves single-fact retrieval, not merely that the model accepts long input, and not robust long-context understanding. Bounded RULER's 'half at 32K' finding to the 17 models it evaluated and flagged it as a 2024 snapshot whose exact fraction moves as models improve. Added the harder-benchmark progression with primary sources: NoLiMa (ICML 2025), which removes the lexical-overlap shortcut, and Sequential-NIAH (EMNLP 2025), which requires ordered needles. Framed the entry around vanilla NIAH within a larger test family, and added a context-rot cross-reference for the length axis.

Perplexity and Gemini now cite this entry for the definition query, reproducing the three-way distinction that citation, mention, and link are separate signals (grounding, naming, and a clickable URL). Two of five tested engines now cite it directly. ChatGPT and Claude cited third-party sources instead this time, so the term draws different sources on different engines.

Claude now cites this entry, reproducing the five-axis diagnostic framework (edge or location, cache state, path pattern, user agent plus referrer, and non-scraper user agent) and the confirmed-versus-candidate classification. Four of five tested engines now cite it directly; ChatGPT, Perplexity, and Gemini already did, each reproducing the same five axes. Microsoft Copilot did not cite it this time.

Perplexity now cites this entry, joining ChatGPT; two of five tested engines cite it directly. The answer reproduced the grounding-metadata framing (grounding chunks, grounding supports, and web search queries) and the distinction from Google AI Overviews and AI Mode. Claude, Copilot, and Gemini itself did not cite it this time.

ChatGPT now cites this entry directly for the definition query, listing it among its sources and reproducing its caveat that AI citation metrics are practitioner-created rather than standardized by peer-reviewed research; the answer closed by recommending per-engine, per-query tracking over a single blended visibility score. Two of five tested engines now cite it directly (ChatGPT and Perplexity).

ChatGPT and Perplexity both now cite this entry directly; with Claude, Copilot, and Gemini already citing it, all five tested engines have now cited this entry at least once. ChatGPT listed two of this glossary's pages as its top sources and reproduced the lock-down-the-denominator guidance; Perplexity cited the 40-of-100 worked example inline.

Microsoft Copilot now cites this entry (second in its references, after Brave's own documentation), reproducing the surfaces framing (AI Answers, Ask Brave, Featured Snippets, AI-powered descriptions) and the independent-index point that Brave citations reflect its own crawl and ranking decisions. Three of five tested engines now cite it directly.

Claude now cites this entry alongside the primary artifacts (the arXiv paper, GitHub, Hugging Face, OpenReview), including the comparison that the retrieval-ranking baseline outperformed the best C-SEO method by 7.6 times. Three of five tested engines now cite it directly, and ChatGPT cites it partially.

Microsoft Copilot now cites this entry first in its references, reproducing the six decisions a protocol locks down (query design, cadence, engine coverage, recording schema, disambiguation rules, signal thresholds) and the three failure modes of unstructured testing. With Copilot added, all five tested engines have now cited this entry at least once.

ChatGPT now cites this entry directly, where it previously appeared only in the folded sources list: the answer led with this page, and the rate-of-change versus snapshot framing plus the momentum-indicator caveat (velocity measures observed output, not the mechanism causing it) were cited inline. Four of five tested engines now cite it directly.

Microsoft Copilot now cites this entry first in its references, reproducing the three citation surfaces (WebSearch inline citations, DeepSearch reasoning traces, and structured citation fields in the xAI API) along with the native X integration and variable citation slot distinctives. Three of five tested engines now cite it directly.

Claude now cites this entry, opening its answer with a direct no and reproducing the Aggarwal et al. 2023 negative result (the only tested method to fall below baseline, at roughly 8 percent worse). Two of five tested engines now cite it directly.

Claude now cites this entry, reproducing the two-tier citation model: licensed publishers receive linked citations under the December 2025 content deals while other sources may appear unlinked, and social posts function as first-class sources alongside web pages. Two of five tested engines now cite it directly.

Microsoft Copilot now cites this entry first in its references, attributing the headline evidence row to it (position-adjusted word count 27.2 versus a 19.3 baseline, about a 41 percent relative gain, the largest measured lift among the nine methods tested). Four of five tested engines now cite it directly.

ChatGPT now cites this entry directly, where it previously appeared only in the folded sources list; it joins Perplexity and Microsoft Copilot, so three of five tested engines cite it directly. Both ChatGPT and Perplexity reproduced the retrieval, grounding, and attribution evaluation framing. Claude and Gemini did not cite it this time.

First confirmed AI-search citation for this entry: Microsoft Copilot lists it in its references for the definition query, alongside the 40 to 70 word extractable-answer framing. One of five tested engines now cites it; ChatGPT, Perplexity, Claude, and Gemini did not cite it this time.

Claude now cites this entry directly, reproducing the precision and recall asymmetry and the Liu, Zhang, and Liang (EMNLP 2023) benchmark (74.5 percent precision, 51.5 percent recall). With Claude added, all five tested engines have now cited this entry at least once across our probe rounds. This round Copilot placed it first in its references and ChatGPT cited it among its sources.

Perplexity now cites this entry directly (first in its sources), reproducing our framing that grounding reduces but does not eliminate hallucinations, since a grounded model can still misread a source or cite one that does not fully support the claim. It joins ChatGPT, so two of five tested engines cite it directly. Claude, Copilot, and Gemini did not cite it this time.

Microsoft Copilot now cites this entry, placing our page at the top of its sources and anchoring to the Aggarwal et al. 2023 result (PAWC 24.7 versus a 19.3 baseline, about 28 percent relative gain on citation visibility). Three of five tested engines now cite it directly, joining ChatGPT and Claude; Perplexity and Gemini did not cite it this time.