AI Search Ranking Factors: What We Know From Testing 450+ Campaigns

We’ve run GEO (Generative Engine Optimization) tests across 450+ client campaigns over the past 18 months. That’s a significant dataset — enough to move from speculation to patterns, from anecdote to evidence. What we’ve found challenges a lot of the conventional wisdom circulating about what makes content rank in AI search. Some factors matter far more than expected. Others that the SEO community obsesses over barely register in AI citation behavior. This is what we know.

How We Collected This Data

Across 450+ GEO campaigns, we tracked citation frequency, answer inclusion rate, source prominence, and competitive citation share across four major AI platforms: Google AI Overviews, Perplexity, ChatGPT (web search mode), and Bing Copilot. We tested controlled content variations — same topic, different structure, different authority signals — to isolate which variables drove AI citation behavior. The findings below are based on statistically significant patterns observed across multiple industries and query categories.

Finding #1: Source Authority Is the Dominant Ranking Signal

Across every platform and category we tested, the single strongest predictor of AI search citation was the overall domain authority of the source. This aligns with what we’d expect — AI models rely heavily on their underlying training data and real-time retrieval, both of which weight authoritative sources.

What “Authority” Actually Means for AI Search

In traditional SEO, domain authority is a proxy metric. In AI search, the relevant authority signals are more specific: editorial reputation (does the domain have a track record of accurate, cited, peer-reviewed or expert-reviewed content?), topical depth (does the domain consistently cover this subject area, not just occasionally?), and external reference density (how many credible sources link to or cite this domain?).

We found that domains with consistent topical authority in a specific vertical were cited 3.4× more frequently than generalist domains covering the same queries, even when the generalist domain had higher overall domain authority scores. Topical authority beats general authority in AI search citation behavior.

Finding #2: Content Structure Has Outsized Impact

AI models need to extract information efficiently. Content that is structured for extraction — with clear headings, defined claims, enumerated facts, and explicit answers to specific questions — is cited significantly more often than narrative prose covering the same information.

The Structure Elements That Drive Citations

From our testing, the following structural elements showed the strongest correlation with AI citation inclusion:

• Direct question-and-answer formatting: Pages that explicitly state a question and immediately answer it (FAQ sections, Q&A formats) were cited 2.7× more often for informational queries than narrative content with the same information buried in paragraphs.
• Statistical claims with attribution: Content containing specific, attributed statistics (not rounded estimates) was preferred by AI models, particularly Perplexity and Bing Copilot. “Approximately 70%” performs worse than “68%, per [Source] 2025 study.”
• Numbered lists and enumerated processes: For how-to and process queries, numbered list formats were cited 2.1× more often than prose descriptions of the same process.
• Definition sections: Content that clearly defines key terms at the beginning of topical coverage performed significantly better for queries asking AI to explain concepts.

Finding #3: Freshness Matters — But Not Equally Across Query Types

Content freshness — how recently it was published or updated — showed a strong positive correlation with AI citation frequency for time-sensitive queries (news, market data, technology updates, regulatory changes). For evergreen informational queries (definitions, processes, frameworks), freshness showed minimal impact and in some cases older, more authoritative content outperformed newer content.

The Freshness-Authority Trade-off

We observed a consistent pattern: for evergreen queries, a 3-year-old article from a high-authority domain outperformed a 3-month-old article from a lower-authority domain in AI citation frequency. For news-adjacent queries, the reverse was true — recency consistently outweighed authority. The implication: your freshness update strategy should be targeted at time-sensitive content, not applied uniformly across your site.

Finding #4: E-E-A-T Signals Are Directly Detectable in AI Behavior

Google’s E-E-A-T framework (Experience, Expertise, Authoritativeness, Trustworthiness) is generally discussed in the context of traditional search ranking. Our data suggests AI models are also applying E-E-A-T-like weighting — and several specific signals drove measurable differences in citation behavior.

Author Credentials and Bylines

Content with bylined authors who have verifiable credentials in the relevant field was cited significantly more often on Google AI Overviews and Perplexity than unattributed content or content attributed to generic “staff” bylines. The signal appears to be domain-specific: a bylined MD drives citation lift for health content; a bylined CFP drives citation lift for personal finance content. Unrelated credentials don’t help.

First-Person Experience Language

Content that explicitly signals direct experience — “In our testing,” “Based on 450 campaigns we analyzed,” “From working with 200+ clients” — was cited more frequently than content making the same claims without experience framing. This is a direct E-E-A-T-aligned signal that AI models appear to weight positively.

Finding #5: External Citations Within Your Content Drive AI Citations of Your Content

One of the more counterintuitive findings: content that cited external high-quality sources was itself cited more often in AI answers than content making unsupported claims — even when the underlying assertions were equivalent. AI models appear to validate content quality partially by checking whether the content itself references credible sources.

We saw a 38% lift in AI citation frequency for content that cited three or more external authoritative sources versus equivalent content with no external citations. This suggests that the well-established journalistic practice of sourcing claims isn’t just good practice — it’s a measurable AI search ranking signal.

Finding #6: Platforms Have Different Citation Preferences

Not all AI search platforms cite content the same way. Understanding platform-specific citation behavior is critical for targeted GEO optimization.

Google AI Overviews

Strong preference for content already ranking in the top 10 organic results. Our data showed that content not ranking on page one organically was cited in AI Overviews less than 8% of the time. Conclusion: for Google AI Overviews specifically, traditional SEO ranking is a prerequisite, not an alternative.

Perplexity

Perplexity shows the most diverse source selection, with greater willingness to cite non-top-10-ranking content. We observed that highly structured content with explicit sourcing and clear factual claims performed particularly well on Perplexity regardless of organic rank. Perplexity also weighted freshness more heavily than other platforms across all query types.

ChatGPT (Web Search Mode)

ChatGPT’s citation behavior was the most variable across our testing. It showed strong preference for established news and media domains for factual queries, and for specialist/niche authority sites for technical queries. For informational content, ChatGPT frequently cited content that would rank poorly in traditional organic search but came from domain-specific high-authority sources.

Bing Copilot

Bing Copilot showed the strongest correlation with traditional Bing organic rankings — more than any other platform. If you’re targeting Copilot citations, traditional Bing SEO (which shares many signals with Google but has important differences in link signal weighting) is the most efficient path.

Finding #7: Long-Form Depth Outperforms Shallow Coverage

In a world of AI content generation, thin content is collapsing in value faster than ever. Our data showed a strong positive correlation between content depth (measured by semantic coverage of the topic cluster, not raw word count) and AI citation frequency.

Pages that covered a topic comprehensively — addressing related questions, providing context, covering objections, offering nuanced analysis — were cited 2.9× more often than pages targeting the same primary keyword with minimal depth. Comprehensive coverage signals to AI models that this is the canonical resource on the topic.

Finding #8: Schema Markup Shows Limited Direct Impact, But Significant Indirect Impact

We tested schema markup extensively — adding FAQ schema, HowTo schema, and Article schema to content and measuring citation frequency changes. The direct citation lift from schema was modest (approximately 12% improvement on average). However, we observed significant indirect benefits: schema implementation correlated with richer search result appearances, which correlated with higher organic CTR, which correlated with stronger overall organic ranking, which — per Finding #5 for Google AI Overviews — correlated with AI citation frequency. Schema’s GEO value is real but indirect.

Finding #9: Brand Mentions Across the Web Amplify AI Citation Probability

Brands that appeared frequently in discussions, reviews, and references across the wider web — not just on their own domains — were cited significantly more often in AI search answers. This is the GEO equivalent of traditional link building: earning mentions across authoritative third-party sources signals to AI models that your brand is the recognized answer to queries in your space.

We saw a 47% correlation between growth in non-linked brand mentions across credible third-party sites and improvement in AI citation frequency for branded and semi-branded queries. The implication: your PR and content distribution strategy is now an AI search ranking strategy.

The Integrated GEO Ranking Model

Based on this data, we’ve developed an integrated GEO ranking model with five primary signal categories, ordered by measured impact weight:

1. Topical domain authority (30%): Sustained depth of coverage in the specific topic area
2. Content structure and extractability (25%): How well the content is formatted for AI extraction
3. E-E-A-T signals (20%): Author credentials, first-person experience markers, external citations within the content
4. Organic search ranking (15%): Particularly for Google AI Overviews; less critical for Perplexity and ChatGPT
5. External brand mention density (10%): Third-party references and mentions across the web

These weights aren’t universal — they shift by platform, query type, and industry. But as a starting framework for prioritizing your GEO efforts, this model is directionally accurate based on our campaign data.

FAQ: AI Search Ranking Factors

Do traditional SEO rankings still matter for AI search?

Yes, particularly for Google AI Overviews, where organic page-one ranking is strongly correlated with AI citation. For Perplexity and ChatGPT, traditional rankings matter less — content authority and structure can compensate for lower organic rankings. For Bing Copilot, Bing-specific rankings are the strongest predictor of citation frequency.

How quickly do AI search rankings respond to content changes?

Faster than traditional search in most cases, but it varies by platform. Perplexity and ChatGPT’s web search mode can reflect new content within days if the content is indexed and from an authoritative source. Google AI Overviews typically lag organic ranking changes by weeks. Don’t expect instant results, but the feedback loops are tighter than traditional SEO for most platforms.

Is there such a thing as negative GEO signals?

Yes. We observed consistent patterns where certain content characteristics negatively correlated with AI citation: content with thin factual depth, content that made claims without supporting sources, pages with significant factual errors (which AI models appear to detect or at least discount), and content that was heavily promotional rather than informational. AI models appear to apply quality thresholds that function similarly to traditional search quality scoring.

Do backlinks matter for AI search ranking?

Indirectly, yes. Backlinks from authoritative sources contribute to your domain’s overall authority, which is the strongest GEO ranking signal we identified. Backlinks also drive higher organic rankings for Google, which correlates with Google AI Overviews citation. However, a content piece with excellent structure and strong E-E-A-T signals from a well-linked domain will outperform a poorly structured piece from the same domain — link equity doesn’t compensate for poor content quality in AI search the way it sometimes does in traditional search.

How does AI search ranking differ between B2B and B2C queries?

B2B queries in technical domains showed stronger weighting toward author expertise signals and external citation quality within the content. B2C queries showed stronger weighting toward experiential language and review-based signals. The E-E-A-T dimension that’s most relevant differs by query type and audience — B2B audiences (and the AI models optimized for them) weight expertise credentials; B2C audiences weight authentic experience.

Will these ranking factors remain stable as AI models update?

The broad factors — authority, structure, E-E-A-T, freshness — are likely to remain relevant because they’re grounded in fundamental content quality signals. Specific weights and platform-specific behaviors will shift as models update. The brands that will consistently win in AI search are those with the measurement infrastructure to detect those shifts quickly and the content systems to respond. That’s a capabilities advantage, not a tactics advantage.