AI search does not simply reproduce a traditional search results page.
This report combines two original Rankability studies. The first tracks demand across 3,751 AI, SEO, AEO, GEO, and AI-agent keywords over 48 months, from June 2022 through May 2026. The second analyzes 1,645 AI citation observations and 916 traditional search observations across the same 31 topics.
For the citation study, we crawled the cited pages and tested their headings, titles, content, links, authority, structure, freshness, and other characteristics. Together, the studies provide 39 findings about how AI search demand is changing and which page characteristics appear alongside AI visibility.
How to read this research: Search volume is a directional proxy for demand, not a direct measure of product usage or market share. The citation findings are observational relationships, not proof that changing one page element will cause an AI platform to cite it. Denominators vary because the two studies and their page-level, query–page, cross-channel, rendered-HTML, and third-party-data analyses answer different questions.
AI search demand over 48 months
Rankability analyzed a fixed panel of 3,751 terms using Google Keyword Planner as the primary demand source. Monthly values were lightly smoothed where needed to reduce bucketing noise, and indexed comparisons rebased themes to a shared starting point.
| Finding | What the demand data showed |
|---|---|
| General AI search demand grew 3.6× | Annualized search demand increased from about 91 million in the 12 months ending May 2023 to about 327 million in the 12 months ending May 2026—an increase of roughly 260%. |
| Interest in SEO terms declined after peaking | Demand climbed into mid-2025, then fell about 30% over the following year. It was the first sustained decline in the four-year panel rather than a short seasonal dip. |
| AEO and GEO became named categories | Search volume was close to zero in 2023, rose quickly through 2024 and 2025, and then leveled off in early 2026. GEO carried roughly twice the search volume of AEO, although both remained small compared with established search terminology. |
| Interest in AI agents became more volatile | Agent-related demand accelerated through 2024 and 2025 before becoming choppier, consistent with attention moving from broad category discovery toward implementation questions. |
The clearest directional change is the widening gap between general AI and SEO demand. AI demand reached 326.9 million in the 12 months ending May 2026, while SEO demand fell from its 2025 peak to 33.4 million.
The smaller categories grew quickly without approaching the scale of general AI or SEO. Looking at the five themes together keeps rapid percentage growth in perspective.
AI-agent demand produced the sharpest sustained increase among the emerging categories, rising from 38,900 to 843,200 over three years.
These findings do not mean traditional search has disappeared. They show attention expanding toward AI answers while the terminology around AEO, GEO, and AI agents continues to settle. The underlying work still overlaps: publish accessible evidence, establish entities clearly, earn third-party corroboration, and measure conventional rankings and AI visibility separately.
Download the complete 48-month demand report
Traditional search and AI visibility
- Traditional top-three results had a 98.9% AI inclusion rate. Across the broader traditional top 10, 90.0% appeared in AI citations, versus 49.3% of observations ranking outside the top 10. Yet only 44.8% of traditional top-10 observations also reached the AI top 10. Traditional rank remained important after controls for authority, length, coverage, focus, evidence, and novelty: it behaved like a gateway, not a guarantee.
- Most AI top-10 citations were not traditional top-10 results. Of 310 AI top-10 observations, 171—or 55.2%—fell outside the traditional top 10; 108 had no observed traditional result and 63 ranked below the top 10. The channels had only 29.1% exact query–URL overlap, and shared results had a pooled rank correlation of 0.389.
Platform fragmentation and source reuse
-
Pages reused across eight or more queries had a 43.3% AI top-10 rate, versus 15.5% for pages used for one query. Broadly reused pages also appeared on 3.33 platforms on average, suggesting that cross-query usefulness aligned with cross-platform reach.
-
Even the most similar pair of AI platforms shared only 24.1% of cited pages. Brave AI and Claude formed that pair; other platform differences were equally stark. Exclusive query–page citation rates ranged from 3.5% for DeepSeek to 73.5% for ChatGPT, while top-domain shares ranged from 4.0% in Google AI Mode to 12.1% in Brave AI. Citation-pool size varied more than eightfold, and smaller platform samples warrant more caution.
Common page foundations
-
A nonempty title appeared on 99.9% of accessible AI-cited pages. Rankability found one in 1,597 of 1,598 observations, making title presence standard but almost useless for differentiating pages already in the cited cohort.
-
H1 presence was almost identical across search channels. Comparable rendered-DOM evidence showed H1s on 98.1% of traditional assets and 97.8% of AI assets.
-
Internal links appeared on 98.5% of cited assets, with a median of 12. Link count was nearly neutral for citation order, so basic internal connectivity was normal rather than a proven scaling lever.
-
External links appeared on 76.0% of cited assets, but only 4.0% used formal reference-style links. External-link count was weak, offering no evidence that adding more outbound links alone improves citation order.
-
Half of cited assets contained a table. The exact share was 50.5%; topic-relevant tables retained a modest adjusted relationship, but table presence was not a universal requirement.
Titles, H1s, and headings
A cross-channel comparison of 1,422 analyzable AI citation observations and 802 traditional observations found topic-matched titles in 85.9% and 95.1% of each sample, respectively. Title relevance separated raw AI citation performance but not traditional Rankability Tracker Domination rank; the analysis is observational and does not prove that rewriting a title causes visibility gains.
Download the comparison tables, sample denominators, and model outputs.
-
Only 4.8% of analyzed AI observations used the complete query in the title. Partial topic matches accounted for 81.2%. Exact-match titles had a 30.9% raw top-10 rate, but topic relevance—not forced query repetition—is the defensible interpretation.
-
The median title was 57 characters in both AI and traditional results. AI top-10 rates ranged from 15.8% to 22.9% across five length bands with no stable progression, and controlled length relationships were small.
-
Only 22.7% of AI-visible pages used identical title and H1 wording. The traditional rate was 24.2%, and median token overlap was 0.778 in both channels. AI top-10 rates were nearly flat from exact duplication at 20.1% to low overlap at 21.1%; topic agreement mattered more than wording agreement.
-
Only 4.0% of AI observations used the complete query in the H1. Exact-match H1s had a 32.3% raw top-10 rate, partial matches 19.9%, and no-topic H1s 11.6%, but the relationship weakened after broader page-relevance controls.
-
Multiple H1s appeared at almost the same rate in both channels: 11.9% of comparable AI-cited assets and 11.1% of traditional assets. Multiple H1s were neither disqualifying nor an established advantage.
-
H2s appeared in 97.6% of accessible AI citation observations. Raw top-10 rates rose with topic alignment, from 11.2% for no-topic H2s to 32.7% for exact matches, yet adjusted H2 alignment was effectively neutral after other on-page controls.
-
Nearly half of cited pages skipped a heading level. Rankability found skipped levels on 49.5% of 1,233 pages. Pages without skips had a 21.8% raw top-10 rate versus 15.9% for pages with skips, but that difference should not be treated as causal.
Technical accessibility and page signals
-
Meta descriptions appeared on 93.9% of accessible citation observations. Topic-aligned descriptions had stronger raw rates, but the independent relationship disappeared after broader controls, making a relevant description a marker of a well-formed page rather than a proven AI ranking lever.
-
Only 2.7% of analyzable non-homepage URLs used the complete query in the path. Exact or partial topic matches appeared in 86.2%; slug relevance related more clearly to platform breadth than citation order.
-
Most AI citations came from domains that did not mention the query topic. No-topic domains represented 85.4% of analyzable rows, while only 10 observations—0.6% of the full citation dataset—used an exact-match domain. Independent domain matching was effectively neutral.
-
HTML represented 99.5% of successfully recovered AI-cited pages. Rankability classified 1,235 of 1,241 resolved URLs as usable HTML; only six were PDFs, too few for a responsible format-performance comparison. Thirty-two requested URLs remained unresolved and were reported separately.
-
Source HTML contained usable content for 92.5% of recovered cited pages. Roughly one in 13, or 7.5%, required browser rendering, and another 88 source-accessible pages exposed an H1 only after rendering. Citation does not prove the AI platform rendered the page itself; it may have relied on a search index or another retrieval layer.
Content depth, focus, and evidence
-
The median AI-cited asset contained 2,681 words, but length was nearly neutral. The middle 50% ranged from about 1,430 to 4,305 words, raw top-10 rates stayed between 17.0% and 20.7% across six buckets, and correlation with citation position was −0.014. Longer pages retained no independent advantage after coverage controls.
-
The complete query phrase was absent from 94.3% of body-content observations. Exact-phrase pages had a higher raw top-10 rate, but they were also more aligned overall; exact keyword density was neutral after broader controls.
-
High semantic-coverage pages had a 27.6% AI top-10 rate, versus 14.5% for low-coverage pages. They also appeared on 2.16 platforms on average, compared with 1.28. Coverage remained associated with citation order and platform reach after broad controls and when each query was removed in turn.
- Pages introducing the topic within 100 words had a 22.1% top-10 rate, versus 8.7% otherwise. Their median citation position was 24 rather than 36, and they appeared on 1.85 platforms on average rather than 1.10.
-
The strongest topical passage began around word 865 at the median, about 36% through the page. Its absolute position was nearly neutral, and direct-answer syntax did not independently outperform other passages. Introducing the overall topic early mattered more than moving one dense passage.
-
Topical coverage mattered more than radical wording novelty. High-coverage pages posted similar top-10 rates whether novelty was high or low—26.8% and 24.7%—while low-coverage groups reached only 13.5% and 10.6%.
-
The highest factual-specificity quartile had a 21.5% top-10 rate, versus 17.0% in the lowest quartile. High-specificity pages appeared on 1.98 platforms on average compared with 1.39; named-entity breadth was the clearest component, with price evidence and methodology language showing modest signals.
-
First-party evidence appeared on 45.9% of pages that ranked top 10 in both channels. At least one threshold-level unique fact appeared on 79.9% of analyzed query–pages, although observational detection cannot prove evidence alone caused visibility.
-
Listicles and comparison pages represented 63.6% of cited assets, but 24 of the 31 queries contained “best” or “top.” That query mix explains much of the apparent page-type preference. Intent alignment was more useful: highly aligned pages had a 20.4% raw top-10 rate versus 8.3% for low-alignment pages, while format became neutral after broader relevance controls.
Links, authority, schema, and provenance
- Nearly three-quarters of AI-cited pages had zero page-level referring domains. The exact share was 72.8%, and the median cited page had zero. This shows backlinks were not a citation prerequisite, not that links never matter.
-
Median domain rating was 63, but one quarter of cited assets came from domains below DR 40. DR correlation with AI citation position was 0.006, so domain authority offered little explanation for ordering within this already-cited cohort.
-
No schema type produced a consistently favorable independent relationship. Article schema appeared on 58.4% of cited assets, Image schema on 62.9%, and FAQ schema on 32.7%. Schema can clarify meaning and eligibility without acting as a universal citation boost.
-
AI and traditional results shared the same 61-day median age. Although 80.5% of reliably dated AI rows were less than six months old, freshness did not separate the channels in this sample and should not be read as a simple recency rule.
-
AI-only pages had a 72.9% AI-or-mixed detector rate, compared with 43.7% for traditional-only pages. AI-like classification did not predict better AI citation position, and detector labels are classifications—not verified authorship.
What the data suggests you should prioritize
Start with technical accessibility: return usable HTML, provide a descriptive title, and use a clear heading structure.
Then focus on the factors with the strongest and most consistent relationships:
- Match the query’s real intent.
- Introduce the topic clearly near the beginning.
- Cover the expected concepts and entities thoroughly.
- Add specific facts, examples, prices, tests, methods, or first-party evidence.
- Measure traditional and AI visibility separately.
- Track the AI platforms that matter to your audience instead of relying on one blended total.
Do not replace those fundamentals with keyword repetition, forced exact-match wording, arbitrary word counts, schema volume, or backlink targets.
Track your AI search visibility
Industry benchmarks reveal patterns. Your own topics reveal where to act.
Rankability Tracker monitors your brand’s AI search visibility and citations across the platforms your audience uses. Traditional rankings, local visibility, and supported video-search surfaces provide additional context in the same client workspace.
FAQ about additional page features
Do FAQ, summary, and table-of-contents blocks help AI citations?
FAQ content appeared on 63.8% of cited assets, summaries on 26.7%, and tables of contents on 11.4%, but none showed a stable independent advantage after broader controls. Add these blocks when they improve comprehension or navigation, not merely to target citations.
Do images and alt text help AI citations?
Meaningful images appeared on 84.8% of cited assets, and median alt-text coverage was 96.3%; however, alt-text coverage did not show a clear independent ranking advantage. Use informative images and write alt text for accessibility and meaning, without keyword stuffing.
Do ordered and unordered lists help AI citations?
Unordered lists appeared on 84.2% of cited assets and ordered lists on 30.7%, but adjusted associations remained weak. Use semantic lists when information is genuinely sequential or scannable rather than fragmenting prose for AI extraction.
Do video embeds and transcripts help AI citations?
Only 7.3% of cited assets embedded video and 1.5% exposed a transcript; embeds showed no stable rank advantage, while the transcript sample was too small for a confident performance claim. Put substantive video claims and evidence in accessible HTML or a useful transcript.