THE STATE OF
AI-DRIVEN SEO

What This Paper Covers

Start here for a quick plan: what we tested, what we found, and what to read first.

How To Use This Report

Table of Contents

The reading order is intentional: findings first, myth tests second, ML appendix third, action pages last.

About This Study

30-Day Trend vs Previous 30 Days

Chart read: These show whether things got better or worse in the last month compared to the month before -- a quick pulse check, not a substitute for the 90-day comparisons in the main findings.

This dataset includes everything -- brand new articles, established pages, and struggling ones. That mix is what makes these patterns real, not cherry-picked. Recent results uses the 90-day window, while structural review uses Full available warehouse history (2025-10-13 to 2026-03-25). Query relevance uses a separate window: Full available query history (2025-06-12 to 2026-03-24).

AI-session totals on this page represent direct blog traffic only -- visits where someone clicked through from an AI tool to a tracked page.

Study Scope & Proof Standard.

Here's exactly what we measured and the rules we follow for interpreting the data.

This paper covers 477,248,928 impressions, 1,537,659 clicks, 1,672,511 sessions, and 17,693 AI sessions.

AI sessions here represent direct click-through visits to the tracked blog pages, not citation or impression share inside AI tools.

Reporting windows: Last 90 complete days for current performance, Last 30 complete days for short-term momentum, Full available warehouse history (2025-10-13 to 2026-03-25) for structural analysis, and Full available query history (2025-06-12 to 2026-03-24) for ranking-query relevance and query-coverage analysis.

Understanding the Metrics

This paper leads with real search results numbers. We use our composite score for context, not as the main proof.

Finding Tags

Method & Limitations.

The Anatomy of Growing Content.

What do growing pages have in common? And what's different about the ones losing traffic?

We split every page into two groups: gaining traffic vs losing traffic. The clearest difference is age. Growing pages are younger and already a little more visible in search. Length is almost the same.

Word count is almost the same in both groups (1.5K vs 1.5K). Age is the real gap. Growing pages are 18.9% younger (185 vs 228 days). Declining pages aren't necessarily bad -- many still get traffic. They're just older and losing momentum.

The Content Results Curve.

Every piece of content has a life cycle. It peaks around 61-90 days, then gets weaker after about 9 months if nobody keeps it sharp.

Health Score by Content Age (Days)

It takes about 60-90 days for Google to fully discover and rank a new page -- that's when results peaks at 37.2 health. After that, results gradually declines. By 271-365 days, health drops to 29.6 as the content gets older. But old does not mean dead: the 365+ group still averages 35.5 health.

Click Capture by Position Tier.

The higher you rank, the more clicks you get -- and the drop-off is steep.

Weighted CTR by Position Tier

The takeaway: improving a page already on page 1 gets you more clicks faster than trying to rescue a page buried on page 5. Once content drops off the first page, click capture shrinks fast.

The Freshness Multiplier.

This section combines two reads: freshness-window growth/decline ratios and a separate 365+ refreshed-vs-stale lift comparison.

Growth-to-Decline Ratio by Freshness Window

Pages updated 0-30 days ago are too new to read reliably. But at 31-90 days, freshness is the strongest measured growth signal -- 5.43:1 growth-to-decline ratio. Pages untouched for 181-360 days show clearly stale behavior at 3.02:1, while the 361+ bucket spikes to 37.19:1 -- but that's a tiny sample with only 21 declining page, so don't read too much into it.

Separate cohort check (365+ pages): among pages older than a year, the cohort refreshed in the last 30 days jumped from 23 to 37 health (1.6x boost) and from 82 to 4.2K impressions (52x) versus old pages last updated 181-360 days ago.

Chart read: Read 31-90 as the strongest stable freshness-window signal. The 1.6x and 52x cards are a separate 365+ refreshed-vs-stale cohort comparison. The 361+ bar is present because it exists in the local sample, but its 37.19:1 ratio is unstable: 781 growing pages versus only 21 declining.

Reader Engagement and Search Visibility Move Together.

High scroll + high reader engagement = +16.1 health points. Search visibility consistency compounds the effect.

Health Score by Scroll Depth x Reader Engagement Rate

Health by Search Visibility Consistency

Pages where people scroll deeper and stay engaged score 48.1 health. Pages people bounce from score 32. That 21.4-point gap is why we use search visibility as context in this paper.

AI Traffic: A Different Signal.

AI traffic does not line up perfectly with normal Google traffic. Here is the pattern we can actually measure.

AI Traffic Share (%) -- Monthly

Note: The most recent month may reflect incomplete data at the time of this export. Read the trend from completed months.

AI Referral Providers (Local Last-30d Breakdown)

Pages with AI referrals do not look exactly like the pages that win normal search. In this dataset, the high-AI group averages 2.6x impressions but slightly weaker Google positions than the no-AI group. So AI visibility and Google visibility overlap, but they are not the same thing.

These numbers track direct blog traffic only -- visits where someone clicked through from an AI tool. The citation and impression reach across AI platforms is much larger. AI click-throughs are still 1.06% of tracked sessions (17.7K of 1.7M). Growing, but still a small share of measured visits. In this export, the highest AI referral rates show up in commercial intent and the 5K+ length bucket. Compared with Ahrefs' 0.1% typical AI-referral benchmark, this site's tracked AI share is about 10.6x higher.

External benchmark note: Similarweb reports AI referrals are growing fast (+357% YoY) even while remaining small versus search volume.

The Winning Combinations.

The easiest win: target topics where people are ready to buy and where competitors haven't shown up yet.

Average Health by Intent x Competition Level

Buy-ready intent + low competition leads every metric in this site: 36.4 health and 3.8K impressions. Learning + low competition pages average 34.1 health -- still solid, but the buy-ready topics clearly outperform.

Among page-1 content, the strongest word-count bucket is 3500+ words (44.8 health, 2.4K impressions). But longer content only wins when the topic genuinely needs that depth -- padding doesn't help.

The Age-Freshness Matrix.

How old is the content and how recently was it updated? This matrix shows how those two factors combine.

Health Score Heatmap: Age Tier x Freshness Tier

Four key zones:
1. Growth Engine (31-90d age, recently updated): 39.02 health.
2. Refresh Winners (365+, updated within 30d): 36.98 health.
3. Decay Zone (181-365d, untouched 6-12 months): 27.57 health.
4. Survivors (365+, 361+ stale): 32.48 health -- very small sample.

The biggest takeaway from this study: a 1-year-old article that you update can compete with brand-new content (36.98 vs 39.02 health). You do not always need net-new content -- you often need updated content on proven pages.

Captured Traffic Value.

The defensible proxy is clicks x CPC, not impressions x CPC.

Click-Equivalent Value by Intent

How much is your organic traffic worth? We multiply actual clicks x what those clicks would cost in Google Ads (CPC). The click-based value is 0.28% of the impressions x CPC number -- that gap is exactly why you should never use the inflated version as a headline.

AI Model Results.

Neither OpenAI nor Gemini content is universally better. When we compare content of the same age, each model wins in different categories.

Age-Controlled Cohort Health

Once we control for age, Gemini leads some cohorts and OpenAI leads others -- the AI model matters less than how well you edit and publish the content. This dataset does not support a blanket penalty tied only to AI usage.

Note: Gemini was not in use across the full history of this site. Older age tiers may have no Gemini data because the tool was adopted later. Compare within the same age cohort to get a fair read.

AI Content Trend Direction.

This shows how AI-written content moved in our dataset over time. We do not just look at the whole pile of traffic. We also look at what happens per page and per client.

Impressions per Active Page (Monthly)

Chart read: Computed as monthly impressions divided by monthly active content. Each active page moved from 722 impressions in 2025-10 to 1.1K in 2026-03.

AI Sessions (Monthly Totals)

Chart read: From 2025-10 to 2026-03, AI sessions moved from 422 to 6.6K.

From 2025-10 to 2026-03, impressions per active page changed +52.1%, total impressions changed +1482.7%, clicks changed +691.4%, and AI sessions changed +1453.6%.

This fallback keeps the direction signal factual for this export while avoiding fabricated client-level claims.

Keyword Drift & Shift.

You pick a keyword, publish content for it, then Google decides what the page actually ranks for. Here's how often that matches.

How Much Traffic Comes From Your Target Keyword

About half the site (51.5%) gets 25% or more of its traffic from the target keyword. The other half drifted -- Google sends them traffic for different queries. And 11.5% of pages get zero impressions from any query matching their keyword. Drifted pages aren't failing -- they rank for 185 queries on average. The median page gets 25.9% of its traffic from its target keyword. The top 10% get 73%+.

Average Impressions by Keyword Match Share

Keyword match share does not predict impressions, clicks, or health score. Drift is not a problem -- it's a feature. The best-doing pages capture their target keyword plus a wide net of related queries. Pages too tightly matched to one keyword have less total reach.

Drift vs Performance.

Does matching your target keyword predict better performance? We checked impressions, clicks, CTR, and health score.

Average Impressions by Keyword Match Share

Chart read: This shows average impressions per page, grouped by how much of the page's traffic comes from its target keyword. Higher keyword match does not mean more traffic.

Average Health Score by Keyword Match Share

Keyword match share does not predict impressions, clicks, or health score. The Pearson correlations are near zero: impressions -0.039, clicks -0.006, health -0.013. Only CTR shows a slight positive trend (0.033).

Pages with 0% keyword match average only 160.5 impressions. The sweet spot is 5–10% match at 6.5K avg impressions and 40.3 health. Pages tightly locked onto one keyword (75-100%) average only 1.7K impressions. Drift is not a problem -- it's a feature.

(Un)Expected Truth: Search Market Share.

AI is growing fast. But search is still much bigger today. And search intent is not the same as AI intent.

Digital Query Share (Q4 2025)

Intent Split (Google vs ChatGPT)

Hard truth: Google is still in billions per day, while AI is still in millions per day. So AI is growing, but it is not replacing search right now.

Second truth: intent does not match 1:1. Google dominates navigational and transactional demand. ChatGPT is stronger in generative/creative demand. Keep two playbooks: one for search demand capture, and one for AI answer visibility.

11 SEO Beliefs Tested.

We tested common SEO assumptions against our data. Here is what each one turned out to be.

"High Search Volume = More Traffic."

Search-volume numbers from SEO tools are rough demand estimates. When you compare them against actual page traffic, they still fail as reliable traffic forecasts.

Share of Pages Beating Search Volume

The adjusted comparison is the cleaner directional read: 58.59% of pages still beat stored search volume after dividing 90-day impressions by three. The correlation signal between search volume and actual traffic is effectively zero. Search volume tells you more about competition than about the traffic a page will actually earn.

"Content With Flags Is Failing."

Pages with diagnosed issues aren't failing -- they're visible enough to measure. That makes them your best improvement targets.

Health Score by Number of Active Improvement Flags

Pages with 3+ flags materially do better than the zero-flag cohort (45.1 vs 33.1). A page needs to be visible enough for us to even detect problems. A page with a specific measurable problem is often more valuable than a page nobody sees. Flags = opportunity, not failure.

"Longer Content Ranks Better."

In this dataset, the 5K+ bucket has the highest average impressions and query coverage, but the pattern across buckets is uneven and descriptive rather than causal.

Average Impressions by Word Count Bucket

The 5K+ bucket leads on both average impressions and average query count, but the trend is not linear across all word-count tiers. Word count has a moderate relationship with query coverage (correlation signal: 0.4127) and a near-zero relationship with reader engagement time (correlation signal: -0.0565). Only 6.17% of pages have their top query as an exact keyword match, while 60.97% are led by an off-target query. Use depth to cover relevant subtopics, not to chase a fixed word target.

"AI-Generated Content Is Penalized."

Almost all content in this dataset is AI-generated. This dataset does not show a blanket penalty tied only to AI use.

Does Google penalize AI-generated content? Our dataset of 342.3K pieces -- almost all AI-generated across multiple model cohorts -- gives us a clear view. When we compared models within the same age tiers, both stronger and weaker outcomes showed up inside AI-authored cohorts. No single model always wins. The differences come from how well the content was edited and published, not from whether AI wrote it.

"Keyword Difficulty Is a Reliable Signal."

Competition level matters -- but not in the way you'd expect. The growth ratio tells the real story.

LOW competition pages score 34.4 health vs HIGH at 29.6 -- a 4.8-point gap. But the growth ratio is where the real difference shows: LOW competition has a 2.1:1 growth-to-decline ratio vs 1.2:1 for HIGH.

That means LOW competition keywords are 75% more likely to be growing than HIGH competition ones. Competition is more useful as a growth difficulty signal than as a simple yes-or-no ranking rule.

"Fresh Content Does Better."

Updating a great page makes it amazing. Updating a weak page makes it slightly less weak.

Health Score by Freshness Tier x Word Count Tier

At 3500+ words: fresh content (0-30d) scores 38.7 health vs stale (181-360d) at 25.4. Freshness multiplies existing quality. Refreshing a strong, comprehensive page produces dramatic results. The top cell in this matrix is 91-180 x 3500+ at 39.1.

"Publishing More Content = Better Results."

We compared the full portfolio by how much they published in the last 90 days vs their actual results.

Results by Publishing Velocity (content-weighted)

High-velocity publishers (> 2,000 pieces in 90 days) average 30.3 health -- roughly double the zero-publishing group (14). That looks decisive, but the exceptions tell a richer story.

A brand in the outdoor subscription box niche reached 45.4 health and 57.0M impressions with zero recent publishing. That is the single highest health score in the entire portfolio. Meanwhile, a speech therapy app brand published nearly 3.0K pieces in 90 days and reached 45.3 -- almost identical health, plus a 15.14:1 growth ratio.

Publishing velocity helps on average, but a strong dormant library can do better than an active mediocre one. The real lesson: volume is a multiplier of quality, not a replacement for it.

"Higher CPC Keywords = Better Organic."

SEO practitioners often chase high-CPC keywords assuming they signal high organic value. Our data says the opposite.

Health Score by CPC Bucket

The cheapest keywords (<$0.50) reach 27.2 health and 2.3K avg impressions, while the most expensive ($5+) manage just 19.9 health. CPC measures ad-market competition, not organic opportunity.

"Better Reader Engagement = Higher Rankings."

Google has repeatedly denied that dwell time and bounce rate are direct ranking factors. Does our data agree?

High reader engagement + high scroll content reaches 48.3 health at position 14. Content with no engagement data sits at 19.4 health at position 16.4. That is only a 2.4 position gap between the best and worst engagement cohorts.

The correlation signal between reader engagement and position is 0.015 -- essentially zero. Scroll rate and position: 0.056. Both are negligible. Engagement does correlate weakly with health (0.115), but health includes engagement in its formula, so that is partly circular.

Position drives reader engagement opportunity, not the other way around. Higher-ranked pages get more visitors who can engage. This supports Google's public statements: engagement metrics are not direct ranking signals.

"Buy-Ready Intent Wins."

SEO teams often prioritize buy-ready keywords for their assumed conversion value. Does intent type actually predict organic performance?

Health Score by Search Intent

Buy-ready intent leads in health (28.8) and impressions (2.4K), but only by a small margin over commercial (27.1). The gap between transactional and informational is just 6.2 health points.

Learning content has the best growth ratio (1.86:1) -- more content is trending up vs down compared to transactional (1.71:1). Informational also holds the highest CTR (33.8%) and dominates volume at 53.4% of the portfolio.

Notably, unclassified content (64.8K pieces) has a 0.56:1 growth ratio -- declining almost 2x faster than growing. Intent classification itself correlates with performance.

"Consistent Search Visibility = Consistent Growth."

If your content appears in search results every day, it must be growing -- right? This is the most counterintuitive finding in our study.

Health Score by Search Visibility Consistency

The "Very Consistent" group (80-90 days visible) has the best health (47.7) and most impressions (8.3K) -- but the lowest growth ratio among visible tiers (1.09:1). Nearly equal up and down. This content has peaked.

The real growth engine is the Moderate group (30-59 days): 3.59:1 growth ratio -- by far the best in the dataset. This content has proven viability but still has room to climb.

The "Rare" group (under 10 days visible) has the worst overall growth ratio at 0.37:1 -- content declining nearly 3x faster than growing. And 146.1K content pieces (42.5% of the portfolio) have zero visibility at all.

The strongest correlation in the search visibility-growth review is search visibility consistency to health at 0.686 (strong). But visibility predicts where you are, not where you are going. The growth sweet spot is moderate consistency -- enough to prove viability, not so much that you have already plateaued.

Indexing Coverage.

Publishing is only step one. FlyRank submits, monitors, and re-checks pages so strong content does not sit invisible.

Coverage

Indexing Coverage

Chart read: This is the fast read: green is the discoverable library, red is content still waiting for pickup, and yellow is content that should be re-checked because it previously showed signs of search visibility.

How FlyRank Manages Indexing

What FlyRank Actually Does

What Actually Correlates?

This chart summary shows which metrics rise and fall together across 211.1K active pages.

Five Types of Content.

We grouped active content into 5 machine-learned types. Each type needs a different action.

Which Pages Will Grow?

We trained a model on 96.6K pages that were clearly growing or declining. It gets it right about 90% of the time on unseen pages from the same brands. On brands it has never seen before, it still works at 75%. Tested 20 different ways across both methods.

Top Growth Predictors

Chart read: Longer bars = bigger influence on growth. The #1 factor is how many days the page showed up in search results. Pages that show up consistently are the most likely to keep growing.

Does the Model's Confidence Match Reality?

Which Dead Pages Can Come Back?

65.7K pages had zero traffic last month. 59% came back on their own. This model gets it right 99% on unseen pages from the same brands, and 97% on brands it has never seen. Tested 20 different ways — never dropped below 93%.

Top Recovery Predictors

Chart read: Pages that had some impressions in the past 90 days and are not very old are most likely to recover. Old pages with zero history are almost certainly dead.

What Will Improve Next Month?

For 112.0K pages with real traffic, this model predicts which will improve more than 10% next month. It gets it right 95% on unseen pages from the same brands, and 90% on brands it has never seen. Our strongest model — tested 20 different ways.

Top Momentum Predictors

Chart read: Pages already getting good impressions relative to their visible days are most likely to keep improving. This is momentum -- pages on an upward path tend to stay on it.

Refreshing Pages Actually Works.

Held-out test summary: 7 of 9 strata show statistically significant refresh lift.

Median Impression Lift by Segment

Where the Numbers Jump.

Some inputs have weak linear correlation but clear threshold jumps.

Word Count -> Median Impressions

Visibility Consistency -> Median Impressions

Every Decision Tested.

Every grouping used in the playbook was tested for statistical significance on held-out data. All passed at p < 0.001 — the differences are real. Every model was tested two ways: on unseen pages from the same brands, and on completely unseen brands. 10 random splits each.

Priority Actions.

8 things you can do right now, ranked by how strong the signal was in our data. Start at #1.

The Content Refresh Playbook.

A step-by-step process for updating content, backed by the freshness and age data in this study.

AI Traffic Improvement Guide.

AI traffic (from ChatGPT, Gemini, etc.) is real but small. Here's what we know about it -- and what to do.

Pages that attract AI visits have a distinct profile: higher impressions, weaker Google positions, and broader search visibility than pages with no AI referrals. Reality check: AI referrals are still only 1.06% of tracked sessions -- growing, but small.

Important: we can't yet reliably tie AI traffic to revenue at the individual visit level, so we focus on search visibility trends instead of making money claims.

The Quick Wins Checklist.

Actions you can take this week, this month, and on an ongoing basis -- ranked by urgency.

We Do One Thing: AI SEO.

Everything in this paper runs automatically for every FlyRank client. Here is (most probably not) full list.

We Fix What's Not Working.

We don't hand you a report and say good luck. We do the work for you.