Last updated: 15 October 2026
How AI engines choose sources comes down to a layered filter: domain-level trust signals, existing Google rank, content structure, and factual specificity. A study of over 18,000 pages found domain-level factors account for 77% of predictive importance for AI citations, with page-level factors contributing just 23% — meaning who you are matters more than any single page you publish (GrowthX, 2026).
Key Takeaways
- A GrowthX study of 18,000+ pages found that domain-level factors account for 77% of predictive importance in AI citation decisions, versus only 23% for page-level factors (GrowthX, 2026).
- Aether AI found that a page ranking in Google's top 3 is 7.82 times more likely to be cited by AI engines than one ranked 11–30 (GrowthX, 2026).
- ChatGPT retrieves far more pages than it ultimately cites, with most of what it retrieves never actually appearing in its final answer (AirOps Research, 2026).
- Adding statistics to a page improves AI visibility by up to 41%, and citing original sources improves it by 30%, per the Princeton/Georgia Tech/IIT Delhi GEO study (Aggarwal et al., ACM KDD 2026, 2026).
- Perplexity's citations overlap with Google's top 10 only 28.6% of the time, far lower than the roughly 76% overlap Ahrefs found for Google's own AI Overviews (GrowthX, 2026; Ahrefs, 2026).
What are AI engine source signals?
AI engine source signals are the combination of domain reputation, retrieval ranking, content structure, and factual density that large language models use to decide which web pages to pull into a generated answer and attribute as a citation. Unlike a traditional search engine, an AI engine such as ChatGPT, Perplexity or Google's AI Overviews doesn't return a ranked list of ten blue links — it retrieves a shortlist of candidate pages, reads them, and selects a handful of passages to quote or paraphrase.
This retrieval process typically relies on RAG (Retrieval-Augmented Generation) — a technique where a language model fetches real-time documents from a search index or the live web before generating a response, rather than relying solely on what it learned during training. The retrieval step narrows millions of pages to a shortlist; the generation step then decides which of those shortlisted pages earn an actual citation.
That two-stage process explains a counterintuitive finding: the vast majority of pages ChatGPT retrieves are never cited in its final answer (AirOps Research, 2026). Being found by the retrieval system is necessary but nowhere near sufficient — the generation model still has to judge a page worth quoting by name.
What signals judge whether a source is trustworthy or authoritative?
Domain-level trust signals judge whether a source is authoritative, and they dominate the decision. GrowthX's analysis of over 18,000 pages found domain-level factors — things like overall site authority, topical consistency, and historical citation patterns — account for 77% of predictive importance, while page-level factors such as word count or keyword placement contribute only 23% (GrowthX, 2026).
This is a different weighting from classic Google SEO, where page-level relevance has traditionally carried more visible influence through on-page optimisation. For AI engines, the site itself functions almost like a credential: a domain with a long history of being cited, linked to, and corroborated by other independent sources earns a form of trust that a single well-written page cannot buy on its own.
"The pattern that repeats: answers move constantly, sources move slowly. Engines redraft their words week to week, but the underlying set of cited domains is sticky. Winning a citation is hard; once you're in the source set, you tend to stay — which is why starting early compounds." — Lauren Dawkins, Head of Content, Aether AI
Practically, this means UK businesses in data, research, finance and professional services should treat domain-building as a long game. A single optimised blog post won't shift an AI engine's trust model overnight; a consistent publishing history, cross-referenced by other credible sites, will.
How do AI engines determine content relevance to a query?
AI engines determine relevance primarily through semantic matching between the query and passages within a document, not just keyword overlap. Modern retrieval systems use embeddings — numerical representations of meaning — to compare a user's question against candidate documents, surfacing pages whose content addresses the question's intent even if the exact wording differs.
Existing Google rank remains one of the strongest proxies for relevance that AI engines lean on. GrowthX found that a top-3 ranked Google page is 7.82 times more likely to be cited by AI engines than a page ranked 11–30 (GrowthX, 2026). AirOps Research separately found that pages ranking #1 in Google are cited 3.5 times more often by ChatGPT than lower-ranked pages, though the firm is clear that rank alone does not guarantee a citation (AirOps Research, 2026).
That relationship isn't universal across engines, though. Perplexity behaves differently to the rest of the field: its cited URLs land in Google's top 10 only 28.6% of the time, compared with roughly 8% for ChatGPT, Gemini and Copilot (GrowthX, 2026). Perplexity appears to run its own independent retrieval and ranking layer rather than leaning heavily on Google's existing order.
Google's own AI Overviews sit closer to classic search than most people assume. Ahrefs' analysis of 1.9 million citations found 76.10% of AI Overview-cited pages ranked in the top 10 organically, with the single most-cited URL on a page showing a median Google ranking of position 2 (Ahrefs, 2026). However, that overlap is shifting fast: a separate, more recent Ahrefs study of 863,000 keywords and roughly 4 million AI Overview URLs found only 38% of cited pages also rank in the top 10 — down sharply from 76% roughly seven months earlier (Ahrefs, 2026). Google appears to be loosening the coupling between its organic rankings and its AI citations over time.
Do AI engines favour certain types of publishers over others?
AI engines do show a measurable bias toward certain publisher types, favouring sites with established editorial reputations, reference authority, or regulatory status. Wikipedia, government domains (.gov.uk), major news outlets, and large reference sites tend to appear disproportionately often in AI Overviews and chatbot citations, reflecting the same domain-trust weighting described above.
For UK businesses, this has a practical implication: a page hosted on a recognised industry body's site — such as guidance published via the Office for National Statistics (ONS), the Information Commissioner's Office (ICO), or a trade association — carries inherent AI-citation advantages that a standalone commercial blog has to work far harder to earn. It doesn't mean smaller or commercial sites can't be cited; original data, named studies, and specific factual claims can offset some of that structural gap, as the Princeton-led GEO research below demonstrates.
| Publisher type | Typical AI citation advantage | Why |
|---|---|---|
| Government / regulator (.gov.uk) | High | Perceived neutrality, regulatory authority |
| Reference sites (Wikipedia) | High | Breadth, cross-linking, update frequency |
| Established news outlets | Medium–High | Editorial standards, frequent indexing |
| Trade bodies / industry associations | Medium | Domain-specific authority |
| Commercial blogs with original data | Medium (earnable) | Specificity and statistics can offset lower domain trust |
| New or thin commercial sites | Low | Limited citation history, few corroborating links |
How important is freshness when AI engines select sources?
Freshness matters more for time-sensitive queries than for evergreen reference content, and AI engines weight recency as one signal among several rather than an overriding factor. A query about "current" rates, rules, or statistics will push retrieval systems toward recently updated or recently published pages, while a definitional or conceptual query is less sensitive to publication date.
The shifting overlap between AI Overview citations and organic top-10 rankings — from 76% down to 38% within around seven months, per Ahrefs (Ahrefs, 2026) — suggests Google's AI Overviews are increasingly willing to surface newer or differently-ranked content rather than defaulting to the same stable top-10 results. That's consistent with freshness gaining weight as a distinguishing signal, particularly on fast-moving topics like regulation, pricing, or statistics.
For data and research publishers specifically, this creates an ongoing incentive to refresh statistics pages, cite the year a figure was published, and republish updated versions rather than leaving a page static for years.
What role do backlinks and citations play in source selection?
Backlinks and citations function as a trust signal that reinforces domain-level authority, the same factor shown to account for 77% of predictive importance in AI citation decisions (GrowthX, 2026). A page that is widely linked to and referenced by other credible sites builds the kind of cross-corroborated reputation that both Google's ranking algorithm and AI retrieval systems reward.
The Princeton-led GEO study adds a page-level dimension to this picture. Researchers from Princeton University, Georgia Tech, the Allen Institute for AI, and IIT Delhi found that citing sources within an article improved AI visibility by 30%, with an effect of up to 115.1% for lower-ranked pages — meaning the citation boost disproportionately helps pages that start from a weaker position (Aggarwal et al., ACM KDD 2026, 2026). The same study found adding statistics improved visibility by up to 41%, and adding direct quotations improved visibility by 28% (Aggarwal et al., ACM KDD 2026, 2026).
Together, these figures suggest citations work two ways: inbound links build a domain's long-term trust profile, while outbound citations — linking to other credible, named sources within your own content — make an individual page more likely to be selected, especially if that page doesn't yet carry heavyweight domain authority.
"Engines cite pages that answer a question completely in one place, say something specific, and come from a source that looks like it should know. Vague pages summarising other pages don't get cited — they get paraphrased without credit. Specificity is the whole game." — Lauren Dawkins, Head of Content, Aether AI
What common mistakes cause a source to be overlooked by AI engines?
Vague, unsourced, and poorly structured content gets overlooked by AI engines most often, because retrieval and generation systems both reward specificity and penalise ambiguity. A page that summarises general industry knowledge without naming a figure, a regulator, a standard, or a date gives a language model nothing concrete to quote — so it gets paraphrased without attribution rather than cited by name.
Common patterns behind missed citations include:
- No named facts: pages that say "many businesses struggle with compliance" instead of naming the relevant legislation, such as the UK GDPR or the Data Protection Act 2018.
- No statistics or studies cited: content lacking any data point misses the up-to-41% visibility lift the Princeton-led study associated with adding statistics (Aggarwal et al., ACM KDD 2026, 2026).
- Broken heading hierarchy: pages that skip from H1 straight to H3, or use bold text as a pretend heading, make it harder for a retrieval system to isolate where one passage ends and the next begins.
- Thin domain history: a single strong article on an otherwise unestablished domain struggles against the 77% weighting toward domain-level trust (GrowthX, 2026).
- Stale statistics: figures left unlabelled or undated reduce confidence for freshness-sensitive queries.
- No outbound citations: articles that make claims without linking to a primary source miss the 30% visibility improvement tied to citing sources (Aggarwal et al., ACM KDD 2026, 2026).
How can a UK business check if its site is being used as an AI source?
A UK business can check AI source usage by directly querying ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini and Microsoft Copilot with the questions its target customers are likely to ask, then noting whether its domain appears in the citations. This manual spot-checking works but doesn't scale — tracking the same set of queries across six engines, repeated weekly as answers shift, quickly becomes impractical by hand.
Aether AI's platform runs citation tracking across all six of those engines — ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini and Copilot — alongside keyword and competitor tracking and Google Search Console integration, so a business can see which of its pages are actually being cited, by which engine, and for which query, rather than guessing. A free AI-visibility audit at Aether AI's /audit gives any UK business a starting snapshot before committing to ongoing tracking.
What steps improve the chances of being cited in AI-generated answers?
A company improves its AI citation chances by combining domain-level credibility building with page-level specificity, since both factors shape the final decision, even at an uneven 77/23 split (GrowthX, 2026). No single tactic guarantees a citation, but the research points to a consistent set of practices that measurably move the odds.
Aether AI's own operational data illustrates the compounding effect of consistent publishing. Across the four brands it currently writes and publishes for — spanning security, facilities software, branding and the platform itself — Aether AI published 281 articles in the 30 days to September 2026, a volume consistent with the domain-level trust-building pattern GrowthX's research associates with stronger AI citation performance, and additionally reflecting the practical output needed to sustain it.
How does AI source selection differ from Google search ranking?
AI source selection differs from Google search ranking by weighting domain-level trust and extractable specificity far more heavily than traditional on-page SEO signals like keyword density or meta tag optimisation. Google's organic algorithm has historically rewarded strong page-level optimisation even on newer domains; AI citation systems make that much harder, since 77% of what predicts a citation sits at the domain level rather than the page (GrowthX, 2026).
The two systems are also decreasingly coupled. Ahrefs found 76.10% overlap between Google's top 10 and AI Overview citations in one large-scale study (Ahrefs, 2026), but a more recent study from the same firm found that overlap had fallen to just 38% (Ahrefs, 2026). Perplexity shows the widest divergence from classic rankings, with only 28.6% of its citations appearing in Google's top 10 at all (GrowthX, 2026). Ranking #1 on Google remains a strong advantage — but it is no longer a reliable proxy for AI visibility on its own.
| Factor | Traditional Google ranking | AI engine citation |
|---|---|---|
| Primary weighting | Page + domain signals, roughly balanced | Domain-level factors, ~77% of predictive importance |
| Top-3 Google rank effect | Strong ranking advantage | 7.82x more likely to be cited (GrowthX) |
| Overlap with top 10 | By definition, 100% | 28.6% (Perplexity) to 76.1% (Google AI Overviews) |
| Content format reward | Keyword optimisation, backlinks | Specificity, statistics, named sources, citations |
| Volatility | Relatively stable over months | Shifting fast — AI Overview overlap fell from 76% to 38% in ~7 months |
Your AI source-selection checklist
- Audit your domain's existing citation footprint across ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini and Copilot before writing new content.
- Name specific statistics, regulators, standards and dates in every article rather than general industry claims.
- Link outbound to primary sources — studies, regulators, standards bodies — to capture the visibility lift associated with citing sources.
- Structure every page with a clean H1 → H2 → H3 hierarchy and no skipped levels.
- Publish consistently rather than relying on a single flagship page to carry domain trust.
- Refresh statistics-led pages regularly and label the year each figure applies to.
- Track citation performance over time rather than assuming a single audit is permanent, given how fast AI Overview overlap has shifted.
FAQ
How do AI engines decide which sources to cite?
AI engines decide which sources to cite by combining domain-level trust signals, existing Google ranking position, and page-level specificity such as statistics and named citations. Domain-level factors account for 77% of predictive importance, according to a GrowthX study of over 18,000 pages (GrowthX, 2026).
Does ranking #1 on Google guarantee an AI citation?
No, ranking #1 on Google does not guarantee an AI citation, though it significantly improves the odds. AirOps Research found pages ranking #1 are cited 3.5 times more often by ChatGPT than lower-ranked pages, but explicitly notes rank alone does not guarantee citation (AirOps Research, 2026).
What is Retrieval-Augmented Generation and why does it matter for sources?
Retrieval-Augmented Generation (RAG) is the technique AI engines use to fetch live web documents before generating an answer, rather than relying only on training data. It explains why most of the pages ChatGPT retrieves are never actually cited in the final answer (AirOps Research, 2026).
Why does Perplexity cite different sources to ChatGPT or Gemini?
Perplexity runs its own independent retrieval system rather than mirroring Google's rankings closely. Its cited URLs land in Google's top 10 only 28.6% of the time, compared with roughly 8% for ChatGPT, Gemini and Copilot (GrowthX, 2026).
Do statistics and quotes really improve AI citation chances?
Yes, statistics and quotes measurably improve AI citation chances. The Princeton/Georgia Tech/IIT Delhi GEO study found adding statistics improved AI visibility by up to 41%, citing sources improved it by 30%, and adding quotations improved it by 28% (Aggarwal et al., ACM KDD 2026, 2026).
Is Google's AI Overview citation behaviour stable over time?
No, Google's AI Overview citation behaviour has shifted substantially in recent months. Ahrefs found overlap between AI Overview citations and Google's top 10 organic results fell from 76% to 38% over roughly seven months (Ahrefs, 2026).
How can a business track whether it's being cited across multiple AI engines?
A business can track AI citations by manually querying each engine or by using a dedicated monitoring tool that checks ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini and Copilot simultaneously. Aether AI offers citation tracking across all six engines alongside a free AI-visibility audit at /audit.
Getting cited: how Aether AI helps UK businesses track and improve AI source selection
Understanding how AI engines choose sources is only useful if a business can see where its own pages currently stand — which is exactly the gap Aether AI's platform closes for UK founders, in-house marketers and SEO leads. Rather than guessing whether a domain's trust signals, content structure or citation history are working, Aether AI tracks citation performance directly across the engines readers actually use.
Aether AI's citation tracking covers all six major AI answer engines — ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini and Copilot — alongside keyword and competitor tracking and Google Search Console integration, giving a single dashboard view of where a business is already being cited and where it is being overlooked.
Start with the free AI-visibility audit at Aether AI's /audit to see a baseline citation snapshot, then explore the self-service platform's public pricing to find the tier that matches ongoing tracking and content needs.