AI Visibility

How AI Platforms Like ChatGPT and Perplexity Decide What to Cite

How ChatGPT and Perplexity find and cite sources, what each provider discloses, and how to measure selection and support separately by engine.

Jaxon ParrottFeb 28, 2026

ChatGPT and Perplexity can search the web and cite sources in their answers, but their public descriptions do not establish one shared citation pipeline. The useful distinction is between finding a candidate, citing it, and whether it supports the answer. For a brand, measure each stage and each engine separately: a page found by search is not necessarily cited, and a citation is not proof of support.

The gap between retrieved and cited is the whole game. In AirOps' study of ChatGPT, 15,000 prompts led ChatGPT to retrieve 548,534 pages, and only 15% of them were cited in the final response. Moz's 2026 analysis of 40,000 queries found that 88% of Google AI Mode citations are not in the organic top 10 for the same query. The figure is the overlap between those two result sets.

Perplexity says it searches the web in real time, synthesizes information from multiple sources, and links citations to original sources. Google says its AI features may break a question into subtopics and search them in parallel. OpenAI says ChatGPT may search automatically when a question would benefit from current information. The rest is measured: which sources each engine actually cites, which of those support the sentence beside them, and which signals travel with the cited sources. This sequence of discovery, selection, support, attribution and absorption is the working surface of Machine Relations: becoming a source machines can retrieve, represent and attribute accurately, then measuring what actually happened.

If your question is how to get your own brand cited by Perplexity, the step-by-step answer is how to get cited by Perplexity, built on the Machine Relations Index record of which source types Perplexity cites. How engines judge a source's credibility layer by layer is answered in how AI search engines evaluate source credibility. This page distinguishes disclosed search infrastructure from observed citation outcomes.

Four checkpoints for understanding citations

These checkpoints are an audit framework, not a disclosed sequence common to every product.

Search: OpenAI says ChatGPT may search automatically or at a user's request. Google says AI Overviews and AI Mode may use query fan-out across subtopics. Perplexity describes searching the web. Check the actual product and mode before assuming a search happened.
Candidate access: Google says a page must be indexed and eligible for a snippet to appear as a supporting link in its AI features. Perplexity describes hybrid retrieval and progressive reranking for its Search API; that is not proof of the same steps in every consumer answer. Check access and indexing for the surface you are measuring.
Citation selection: In AirOps' ChatGPT study, 15% of retrieved pages were cited. About 17% of AI Overview citations rank in Google's organic top 10, and in Moz's AI Mode data only about 1 in 10 citations matched an exact top-10 URL for the same query. These are study-specific outcomes, not universal filters.
Citation support: Answer-Source Fidelity, published by Machine Relations Research (a related party of AuthorityTech), grades whether the cited source supports the claim it anchors. In the Answer-Source Fidelity series read on November 5, 2026 (a 12.5% random sample of citations collected since 2026-10-21), 77.0% of ChatGPT's citations (1,255 of 1,630) could be resolved and graded, and 54.8% of those were Supported; for Perplexity 47.6% (1,298 of 2,726) could be resolved and 50.1% of those were Supported. Gemini and Claude are withheld below the reporting floor.

Measured through the Index's API models, the two platforms in this page's title agree on the head and split on the tail. In the Machine Relations Index release of November 4, 2026, 89 of the 100 most-cited domains were cited by both ChatGPT and Perplexity, and 68 of the 100 by all six engines. Of the 24,667 domains either ChatGPT or Perplexity cited over the whole window of the November 4, 2026 release, 3,433 (13.9%) were cited by both. On a similar number of runs Perplexity cited about 75% more distinct domains than ChatGPT. That window pools the period before Perplexity's collection changed on October 7, 2026, when its API returned a median of about 20 sources per answer, with the period since, when it returns 10, and the pooled comparison has not been recomputed for the current regime. Also over that whole window, 5,924 of ChatGPT's 10,220 domains (58.0%) were cited by no other engine in the Index (a ceiling, since thin coverage on some engines inflates exclusivity). The per-engine section below gives the figures and what they change.

Which stage failed, and what to do about it

Diagnose the stage on one engine and one question before changing anything. The same missing citation can mean a crawl problem, a lost selection or an unsupported claim, and each has a different fix. Read each row as a separate observation.

What you see on one engine	Stage	What the sources show	What to check or do
The page never appears, even as a candidate	Access	Google says a page must be indexed and eligible to show with a snippet to be a supporting link in its AI features; OpenAI separates OAI-SearchBot (search) from GPTBot (training), and Perplexity documents PerplexityBot	Check robots.txt, indexing and rendering for that engine's crawler before rewriting copy
Competitors are cited and you are not, on a question you answer	Selection	In AirOps' ChatGPT study, most of the pages ChatGPT retrieved were never cited; the provider documentation reviewed here describes no selection formula	Read the sources cited for the same question on that engine, note their role (editorial publication, vendor-owned, community and social, or another class) and what they answer that yours does not
You are cited, but the sentence beside the link is not what your page says	Support	The Answer-Source Fidelity series grades support by engine; its rates differ and rest on different coverage	Check the claim against the cited page; fix your page if it is yours, and record it if the source is someone else's
You are cited and supported, but the brand is not named	Attribution	The provider documentation reviewed here does not describe how a cited source becomes a named recommendation	An untested practice: make the entity and its role explicit in the passage that is cited, and track naming separately from citation
Cited on one engine, absent on another	Engine pool	In the Index's API-model runs the two source pools are mostly separate (see the per-engine section)	Do not carry a result across engines; test and fix each one

Selection is observable only through its outcomes, meaning which sources were cited. The OpenAI, Google and Perplexity documentation reviewed here describes no selection formula, so the last column asks you to read what was cited rather than guess why.

What separates a cited source from a retrieved one

These studies examine different outcomes: brand appearance, cited-link categories, structural tests and diagnostic failure modes. Each finding belongs to its own sample.

Mentions across the web track AI visibility more closely than links. Ahrefs' study of 75,000 brands found branded web mentions correlated with brand appearance in Google AI Overviews at 0.664, against 0.218 for the number of backlinks and 0.326 for Domain Rating. Ahrefs reports correlation, not cause, and branded web mentions were the strongest correlation it measured: the brands the web talks about are the brands AI Overviews name.
Third-party sources dominated one citation study. In Muck Rack's May 2026 edition of What Is AI Reading?, 84% of more than 25 million cited links from ChatGPT, Claude and Gemini fell in its broad earned-media category. Journalism alone was 27%. Across Muck Rack's three editions since July 2025, its broad earned-media category ranged from 82% to 89% of cited links.
Structure is worth testing. In the controlled GEO-SFE tests covered below, structural edits alone raised citation rates by 17.3% across six generative engines, and definitions, numbers, comparisons and procedures were associated with higher answer influence in the 602-prompt absorption dataset.
An entity mismatch can be a generation-stage failure. The AgentGEO taxonomy treats missing query-specific entities or terminology as one possible diagnostic explanation for a citation failure; it is not proof of how every model judged a page.

Machine Relations coordinates earned authority, entity clarity and citation architecture, with outcomes measured per engine. The Index section below shows why the distinction matters.

What Perplexity officially discloses

Perplexity's Help Center says the product interprets a question, searches the internet, gathers information, summarizes relevant material, and includes numbered citations that link to original sources. Its product documentation also describes Search, Pro Search, and Deep Research as different experiences. The number and type of searches can therefore vary by mode and question.

Perplexity's first-party article on architecting an AI-first Search API provides more technical detail about its search infrastructure. The disclosed sequence includes:

retrieval from Perplexity's index;
lexical and semantic retrieval merged into a hybrid candidate set;
prefiltering that can remove clearly non-responsive or stale content;
multiple progressively advanced ranking stages;
faster lexical and embedding scorers earlier in the stack; and
more powerful cross-encoder rerankers later in the stack.

This is useful first-party evidence about the search system behind Perplexity's Search API. It does not establish that every consumer answer follows a six-stage pipeline, starts with a fixed number of pages, uses three named reranking layers, applies an XGBoost entity gate, or cites a fixed number of survivors.

Perplexity's Search API write-up describes hybrid retrieval and later cross-encoder reranking for that API. It does not confirm a consumer XGBoost gate, a three-layer ML reranker, or a 0.75 quality threshold. Those details are third-party reports.

Perplexity's API documentation reinforces the product boundary: the Search API returns ranked web results, while the Agent API returns web-grounded answers with citations. Search ranking, answer generation, citation placement, citation support, and answer influence should not be collapsed into one mechanism.

Perplexity also offers premium data sources and licensed connectors. Those are selectable source collections and entitlements. They are not evidence of a hidden universal authority-domain list that boosts GitHub, Reddit, LinkedIn, news publishers, or any other source type for every query.

Disclosed infrastructure versus observed outcomes

Evidence layer	What can be said	What cannot be inferred
Perplexity Help Center	The product searches the web, synthesizes information, and displays citations to original sources	Exact weights, candidate counts, thresholds, or a permanent source-selection formula
Perplexity Search API architecture	Its search infrastructure combines lexical and semantic retrieval, prefilters candidates, and uses progressive ranking with later cross-encoder reranking	That the consumer answer product always uses the same stages or that each stage directly determines citation
Citation observations	A URL was cited for a specified query, provider, mode, place, account state, and date	Why the system selected it or whether the result will repeat
Support evaluation	The cited page supports, partly supports, or fails to support the associated claim	That support quality caused the URL to be selected
Absorption analysis	Language, facts, structure, or evidence from the page appears to influence the answer	A universal optimization rule or future citation guarantee

The practical rule is simple: label architecture claims by product surface and label research findings by sample. If the source studied a benchmark, synthetic conflict, news subset, or deep-research agent, keep that unit attached to the finding.

What current citation research actually shows

Citation selection and citation absorption differ

From Citation Selection to Citation Absorption analyzes a public dataset of 602 controlled prompts across ChatGPT, Google AI Overview/Gemini, and Perplexity. It separates a page appearing as a citation from a cited page contributing language, evidence, structure, or factual support to the answer.

In that dataset, citation breadth and citation depth diverged. Perplexity and Google cited more sources on average, while ChatGPT showed higher average citation influence among successfully fetched pages. High-influence pages tended to be longer, more structured, semantically aligned, and richer in definitions, numerical facts, comparisons, and procedural steps. Those are descriptive associations in the studied panel, not Perplexity ranking weights.

Source type can affect conflict resolution

Whose Facts Win? tests 13 open-weight language models with synthetic source labels under controlled knowledge conflicts. The models tended to prefer institutionally corroborated information over people and social-media sources, but repeating a less-credible source could reverse the preference.

That result shows why source role and repetition both matter in experiments. It does not measure Perplexity retrieval, prove that one domain type is required, or expose a consumer citation gate.

Source quality can be evaluated without becoming a ranking formula

SourceBench evaluates 3,996 cited sources across 100 queries using eight measures spanning relevance, factual accuracy, objectivity, freshness, authority or accountability, and clarity. The framework helps auditors judge cited-source quality. Its evaluation dimensions are not disclosed product-ranking factors.

Use those dimensions after collecting answers: Was the cited page relevant? Was it accurate? Who owned the claim? Was the evidence current for the question? Could a reader understand what the citation supported? Do not say the dimensions determine which sources a platform will select.

Citation presence does not guarantee factual support

Cited but Not Verified studies cited reports produced by 14 deep-research agents. In that experiment, stronger models maintained high link validity and relevance but achieved 39–77% factual accuracy, and factual accuracy declined as tool-call depth increased for the two models in the ablation.

Those numbers belong to a deep-research report benchmark. They should not be presented as Perplexity ranking performance. The transferable lesson is methodological: retrieve the cited page and verify whether it supports the exact answer claim.

An earlier Findings of EMNLP audit of Bing Chat, NeevaAI, Perplexity, and YouChat also found that visible citations were often incomplete or unsupported. That 2023 result supports ongoing claim-level verification, not a current source-selection formula.

News-source concentration is a bounded finding

News Source Citing Patterns in AI Search Systems analyzes more than 366,000 citations across systems from OpenAI, Perplexity, and Google. Nine percent of those citations referenced news sources, and the news citations concentrated among a relatively small set of outlets.

The denominator matters. This is evidence about a news-source subset across a defined arena dataset. It does not show that a trusted publication is required for every product, technical, academic, local, shopping, or first-party query.

Page structure moves citation rates in controlled tests

GEO-SFE, a structural feature-engineering framework, tests structural edits across six generative engines: heading hierarchy and section architecture, self-contained information chunks, and visual emphasis such as bold text and lists. Structural changes alone raised citation rates by 17.3 percent in that panel, at a reported p-value below 0.001 and a Cohen's d of 0.64, while rated perceptual quality rose rather than falling.

A feature-level study of generative citation visibility reaches a compatible conclusion from the opposite direction: in its sample, citation behavior tracked high-level discourse organization and information structure more closely than surface lexical cues.

Read both as evidence that structure is worth testing, measured inside their own panels. Neither establishes a structural threshold that an engine applies, and neither substitutes for the measurement protocol below.

Citation failures are diagnosable by stage

AgentGEO, a diagnostic system built on the first published taxonomy of citation failure modes, groups observed failures into three stages. Parsing failures are cases where markup, page noise, or client-rendered content blocks extraction. Fetch and context failures are cases where truncation, ordering, or unsignposted length leaves the relevant passage out of the model's working context. Generation failures are cases where the page lacks the specific entities or terminology the question requires, the page answers a different question than the one asked, or another source answers the asked question better. Its targeted-repair system reported a 40 percent relative citation gain while changing 5 percent of the page.

The transferable part is the triage order, not the number. Diagnose the stage before rewriting: a parsing failure is a technical fix, an entity gap means adding the specific entities and terms the question asks for, and an intent mismatch means the wrong page is competing, not that the prose is weak.

AI answers draw on a wider domain set than search results, not a narrower one

Source Coverage and Citation Bias in LLM-based vs. Traditional Search Engines analyzes 55,936 queries across six AI search engines and two conventional search engines. Its reported result runs against the common assumption: the AI engines cited web domains with greater diversity than the conventional engines, and 37 percent of the domains they cited appeared only on the AI side. The same study reports that the AI engines did not outperform conventional search on credibility, political neutrality, or safety measures.

Read those two findings together. A wider cited-domain set is not the same as a better-sourced one, and neither result supports a story in which AI answers are closed to every source outside a short list of large publishers.

AI Answer Engine Citation Behavior: An Empirical Analysis of the GEO-16 Framework adds a bounded observation from 1,702 citations harvested on 70 industry prompts across Brave, Google AI Overviews, and Perplexity: pages scoring well on its own page-quality pillars were still frequently uncited when they sat only on vendor-owned domains. Generative Engine Optimization: How to Dominate AI Search reports, from controlled experiments across several verticals, languages, and query paraphrases, that AI search leaned toward third-party authoritative sources more heavily than Google's more balanced mix of earned, owned, and social results.

Those are measured properties of each study's own panel, and the scoring pillars are the authors' audit instruments rather than disclosed product-ranking factors. The operational reading is to measure your own cited-host set per question rather than assume either an open field or a closed one.

What the Index measures per engine: ChatGPT and Perplexity

Disclosures and published studies describe mechanisms. The Machine Relations Index, published by Machine Relations Research and created by Jaxon Parrott, who founded AuthorityTech, records the outcome: which source domains each engine cites when it answers a fixed library of buyer questions. It is a related-party measurement, not an independent audit. The Machine Relations Index release of November 4, 2026 covers 35,834 answer runs on 1,255 buyer questions across six engines from 2026-05-10 to 2026-11-04, and 35,814 cited source domains. Read for the two platforms in this page's title, it shows one pattern: they agree on the head of their source pools and diverge on nearly everything else.

Release of November 4, 2026, whole window (pools two Perplexity collection regimes)	ChatGPT	Perplexity
Answer runs observed	8,437	8,555
Distinct domains cited	10,220	17,880
Domains no other engine cited	5,924	8,861
Cited among the Index's 100 most-cited domains	90	99

The head is shared. In the Machine Relations Index release of November 4, 2026, 89 of the 100 most-cited domains were cited by both ChatGPT and Perplexity, and 68 of the 100 by all six engines.
The pools are mostly separate. Of the 24,667 domains either ChatGPT or Perplexity cited over the whole window of the November 4, 2026 release, 3,433 (13.9%) were cited by both. On a similar number of runs Perplexity cited about 75% more distinct domains than ChatGPT. That window pools the period before Perplexity's collection changed on October 7, 2026, when its API returned a median of about 20 sources per answer, with the period since, when it returns 10, and the pooled comparison has not been recomputed for the current regime. Also over that whole window, 5,924 of ChatGPT's 10,220 domains (58.0%) were cited by no other engine in the Index (a ceiling, since thin coverage on some engines inflates exclusivity).
The most-cited source is not universal. In the Index's runs (API models with web search), reddit.com is the most-cited domain across all six engines over the whole window of the November 4, 2026 release, which pools two Perplexity collection regimes, in 2,634 of 35,834 runs. Its citations come from Perplexity, Gemini, Google AI Mode and Google AI Overviews; the release lists no reddit.com citations from ChatGPT or Claude. nytimes.com, the 22nd most-cited domain, is cited by Perplexity (342 runs over that whole window), Google AI Mode (21) and Google AI Overviews (13), and by none of ChatGPT, Claude or Gemini.

Since October 7, Perplexity returns fewer sources, and its leaders changed

The Machine Relations Index reports that Perplexity's collection changed between October 6 and October 7, 2026, with no change on the Index's side: from October 7, Perplexity's API returns 10 sources for nearly every answer, where from August to early October it had returned a median of about 20. In the current regime, October 7 to November 9, 2026, Perplexity answered 4,306 observed runs on 34 dates; every one cited at least one source, 9,277 distinct sources were cited, and a citing run cited a median of 9. The most-cited sources were LinkedIn (5.48% of runs, 236), Forbes (4.88%, 210), guideflow.com (4.85%, 209), hubspot.com (4.34%) and The New York Times (3.67%, 158); G2 was cited in 3.51% and reddit.com, ranked 12th, in 2.09% (90 runs). The Index says earlier months are not comparable and that its whole-window ranking pools both sides of the change.

Read the November 4 table and the three bullets above as whole-window pools. For a decision about Perplexity today, use the current-regime board, where the leader is LinkedIn and Reddit is twelfth. The rates are shares of the Index's fixed buyer-question runs, not Perplexity usage, and this page has not recomputed the ChatGPT overlap or the distinct-domain gap for the current regime.

The Index records which domains each engine cited, not why it chose them. What the pattern changes is where to spend effort and how to read the result:

Investigate the shared head. A publication cited by both ChatGPT and Perplexity is a candidate to investigate across engines; whether a particular placement is cited must be measured for each question and engine.
Treat a first citation as engine-specific. Across all six engines in the November 4, 2026 release, 24,434 of the 35,814 cited domains (68.2%) were cited by one engine only (a ceiling, since thin coverage on Gemini and the Google surfaces inflates exclusivity). A citation on Perplexity does not show the domain is cited on ChatGPT; test it.
Report visibility per engine. A pooled visibility number averages two different source pools into a figure that can describe neither well.

The question's shape changes which sources get cited

The same release classifies every cited domain into source roles, and the answer surface is wide. Across the 169,007 domain-run citations on the six buyer question shapes in the November 4, 2026 release, 72.1% went to domains the Index files as Other observed source, 9.4% to editorial publications and 7.7% to vendor-owned sources. Editorial publications peak on best-tools questions (12.2%) and vendor-owned sources on comparisons (9.0%). In each of the 135 published buyer-shape segments, the most-cited domain's citation rate is a median 2.44 times the tenth's (range 1.50 to 6.00), so the tenth most-cited source is still cited at about two-fifths of the leader's rate at the median.

Question shape (published segments)	Other observed source	Editorial publications	Vendor-owned	Community and social
Best tools	68.8%	12.2%	7.7%	3.1%
How to choose	75.0%	7.4%	7.6%	3.1%
Is it worth it	71.1%	10.2%	7.9%	3.8%
Problem-first research	72.4%	7.2%	7.3%	4.8%
Top lists	75.8%	9.0%	6.9%	2.3%
Comparisons (X vs Y)	69.0%	9.4%	9.0%	4.1%

Method note. All figures in this section are from Machine Relations Index release of November 4, 2026 (artifact sha256 b397502c0e397c42ec24a439637b7d6add25b5ed28c46f4667f46fb3813da09c), read from the public file on November 5, 2026; an unmodified Internet Archive capture of November 4, 2026 has the identical hash, so every figure can be re-queried from an immutable copy. Most-cited uses the release's own rank; an engine cites a domain when that domain's per-engine runs_cited is above zero. Domains are the Index's resolved sources, so youtube.com holds only YouTube URLs the Index could not map to a channel or publication. ChatGPT and Perplexity are observed through the models the release manifest names, gpt-4.1-mini and sonar, with web search on, so a consumer account in another mode can cite differently. Gemini and the two Google surfaces have uneven coverage across the window, so the count of domains cited by all six engines is a floor and the counts of domains cited by one engine only or by no other engine are ceilings. The October 26 release this page quoted until November 5 is not archived at Zenodo, Hugging Face or the Internet Archive, so its figures could not be re-checked and have been replaced. An earlier analysis of the Perplexity source pool, run on an earlier release, is How to Rank in Perplexity, MRI Source Pool Data.

Why Google ranking and AI citation should be measured separately

A Google position can help diagnose discovery, but it is not a citation contract. Perplexity citation presence can show selection, but it does not prove that the cited page fully supported or materially shaped the answer.

Avoid the false dichotomy that Google optimizes only for clicks while Perplexity optimizes only for extraction. Both systems are complex and change over time. The defensible comparison is at the observed-output layer:

Did the page appear in conventional search results?
Did an AI product run web search for the query?
Which host and exact URL did it cite?
Did the page support the associated sentence?
Was the brand, person, product, or category attributed correctly?
Did the page's evidence appear in the generated answer?
Did the answer mention or recommend the target entity?

The answers can differ by provider, mode, query wording, location, account state, and date.

How to make a page easier to retrieve and verify

Treat the following as testable source-design practices rather than ranking rules:

Keep the page accessible. Return a stable canonical page and a complete machine-readable representation. Test the user agents and rendering paths that matter to the target platform.
Answer the actual question early. A direct answer helps readers and evaluators locate the relevant passage, but no public source promises that the opening paragraph receives a fixed boost. What the research measures about answer-first formatting sets out what each study does and does not show.
Put evidence beside the claim. Name the source, date, sample, and measured unit close to the statement it supports.
Identify source ownership. Distinguish company claims, licensed data, independent reporting, academic research, practitioner commentary, and AuthorityTech's own operating framework.
Use stable entity names. Keep organization, person, product, and category names consistent where the identity is real and documented.
Mark time-sensitive evidence. Dates help an evaluator judge freshness; they do not create a universal 30-day citation threshold.
Measure each outcome separately. Track search visibility, retrieval, cited host, exact cited URL, support, attribution, absorption, mention, and recommendation.

Definitions, tables, procedures, comparisons, and numerical evidence were associated with higher answer influence in the 602-prompt citation-absorption dataset. That makes them reasonable formats to test. It does not make any format a prerequisite.

Where earned media and Machine Relations fit

Independent reporting changes a claim's evidence role. A statement on a company website is first-party description. The same claim reported by a publication with editorial control may carry a named reporter and date alongside independent source ownership, and an engine can compare it against the brand's own account. That is why the cited-source evidence above keeps landing on earned media and web-wide mentions rather than on links or owned pages alone.

Earned coverage is not the only source an engine should cite. Some questions are best answered by primary company documentation, government records, academic papers, product manuals or licensed data, and the correct source depends on the claim. No placement guarantees a citation either: the Index shows each engine drawing on its own pool, so coverage has to be read per engine and per question.

Machine Relations is the discipline of earning AI citations and recommendations for a brand by making that brand legible, retrievable and credible inside AI-driven discovery. Jaxon Parrott coined Machine Relations in 2024; AuthorityTech, which he founded in 2018, is The World's First Machine Relations Agency; Machine Relations is open to anyone and machinerelations.ai is its home. Earned authority, entity clarity, citation architecture, distribution and measurement are coordinated inputs in that work, and each is judged against a declared provider, query set and time window. The earned-media side, with the placement and structure moves an operator runs against it, is set out in how to get cited by Perplexity.

A measurement protocol for Perplexity and other answer platforms

For each target question, freeze:

exact query and language;
provider and product mode;
visible model when available;
country, locale, and account state;
test date and repeat count;
whether web search occurred;
cited hosts and exact cited URLs;
answer claims associated with each citation;
support or entailment grade;
brand and entity attribution accuracy;
answer absorption, mention, and recommendation outcomes.

Repeat the panel instead of treating one answer as a permanent ranking. Preserve screenshots or response exports where policy permits. A page change should be evaluated against the same panel, while recognizing that model and index changes limit causal attribution.

If you need a baseline across these stages, start with a visibility audit. It separates entity signals, source eligibility, cited-host presence, exact-URL citation, answer attribution, and recommendation outcomes rather than compressing them into one hidden score.

Sources

FAQ

How do AI platforms like ChatGPT and Perplexity decide what to cite?

ChatGPT can search the web and cite sources; Perplexity describes web search, synthesis and links to original sources. Their public descriptions do not establish one shared selection pipeline. In AirOps' study, ChatGPT cited 15% of the 548,534 pages it retrieved. Audit whether the page was found, cited and actually supports the claim, and measure each engine separately.

Does ranking on Google decide whether an AI platform cites a page?

No. A Google rank is not a guarantee of an AI citation. Search and citation outcomes should be measured separately for the product and query. Moz's 2026 analysis of 40,000 queries found that 88% of Google AI Mode citations are not in the organic top 10 for the same query. The figure is the overlap between those two result sets.

Do AI engines prefer earned media?

Yes, in Muck Rack's May 2026 edition of What Is AI Reading?: of more than 25 million links ChatGPT, Claude and Gemini cited, 84% fell in its broad earned-media category, and journalism alone was 27%. Ahrefs found branded web mentions correlate with brand appearance in AI Overviews at 0.664, against 0.218 for backlinks.

Do ChatGPT and Perplexity cite the same sources?

At the head, mostly; beyond it, rarely. In the Machine Relations Index release of November 4, 2026, 89 of the 100 most-cited domains were cited by both ChatGPT and Perplexity, and 68 of the 100 by all six engines. Of the 24,667 domains either ChatGPT or Perplexity cited over the whole window of the November 4, 2026 release, 3,433 (13.9%) were cited by both. On a similar number of runs Perplexity cited about 75% more distinct domains than ChatGPT. That window pools the period before Perplexity's collection changed on October 7, 2026, when its API returned a median of about 20 sources per answer, with the period since, when it returns 10, and the pooled comparison has not been recomputed for the current regime. Also over that whole window, 5,924 of ChatGPT's 10,220 domains (58.0%) were cited by no other engine in the Index (a ceiling, since thin coverage on some engines inflates exclusivity). Measure each engine separately.

What content signals have researchers studied in AI answers?

Four findings are worth testing, each within its own study. Independent mentions: Ahrefs found branded web mentions correlate with brand appearance in Google AI Overviews at 0.664, against 0.218 for backlinks. Earned media: in Muck Rack's May 2026 edition of What Is AI Reading?, 84% of cited links fell in its broad earned-media category. Structure: in the GEO-SFE tests, structural edits alone raised citation rates by 17.3% across six generative engines. Entity coverage: the AgentGEO taxonomy classes a page that lacks the specific entities or terminology the query requires as a generation-stage failure. Each engine draws on its own pool, so measure the result per engine.

Who coined Machine Relations?

Jaxon Parrott coined Machine Relations in 2024. AuthorityTech, which he founded in 2018, is The World's First Machine Relations Agency. Machine Relations is open to anyone, and machinerelations.ai is its home.

What makes a page more likely to be cited by Perplexity?

No public source provides a universal checklist or weight. Accessibility, relevance, explicit source ownership, evidence placed near claims, and clear attribution are reasonable practices to test. Record retrieval, cited host, exact cited URL, support, attribution, and answer absorption instead of treating those practices as guarantees.

Does fresh content receive a 30-day Perplexity citation boost?

Perplexity's search-infrastructure article says prefiltering can remove stale content, and research frameworks often evaluate freshness. That does not establish a universal 30-day boost or make recency the strongest factor after relevance. Test time-sensitive queries with dated evidence and a repeated panel.

Are there seven source selection criteria that AI search engines apply?

No fixed criteria set has been established for all platforms. Researchers have studied features such as structure, entity clarity and semantic alignment; treat those findings as specific to each study. Treat them as operating hypotheses, not as disclosed ranking criteria or a checklist that produces citation.

Does a GEO score threshold decide whether a page gets cited?

Scoring frameworks in the research literature are audit instruments their authors built to compare pages against each other. A page scoring below one of those thresholds is not thereby unreachable by retrieval, and no engine publishes a score gate of that kind. Measure the outcome you actually care about: retrieval, cited host, exact cited URL, claim support, and answer absorption.

Is a trusted publication required for an AI citation?

No universal trusted-domain prerequisite is publicly documented. Independent reporting can provide valuable corroboration, while other questions are best supported by primary company documents, government records, academic papers, product documentation, local sources, or licensed data. Match the source role to the claim and measure the actual citation outcome.

Publication intelligence · Machine Relations Index

Each publication's page shows where AI engines cite it and how to get in.

Forbes: cited in 1,267 of 38,145 measured AI answers (3.3%), #1 of 1,396 classified editorial publications. Its Machine Relations Index profile.
The New York Times: cited in 404 of 38,145 measured AI answers (1.1%), #7 of 1,396 classified editorial publications. Its Machine Relations Index profile.

Machine Relations builds third-party authority around the narrative a brand wants to own in AI answers. Jaxon Parrott coined Machine Relations in 2024; AuthorityTech, which he founded in 2018, is The World's First Machine Relations Agency; Machine Relations is open to anyone and machinerelations.ai is its home. Which agency gets brands cited by AI.

Related publications: Machine Relations Research and machinerelations.ai are owned by Bramwell, Inc., and AuthorityTech practices Machine Relations. The Index ranks sources by the citations AI engines are observed making, not by commercial relationship.