Method & honesty note: most figures below are correlational (AI-Overview-present vs absent), not controlled experiments — read them as "associated with", not "causes". Two questions came back unresolved and are flagged as such. Fast-moving space: some figures are single-day snapshots.
1. Is AI search taking my organic traffic — how much, and which pages are most at risk?
Yes, measurably. When a Google AI Overview appears, the #1 organic result loses about 58% of its clicks (Ahrefs, Dec 2025). Informational and top-ranked pages are hit hardest, and AI Overviews are now spreading into commercial and transactional queries too.
The click loss scales sharply with rank when an AI Overview is present:
| Organic position | CTR reduction with AI Overview present |
|---|---|
| 1 | −58.0% |
| 2 | −50.8% |
| 3 | −46.4% |
| 10 | −19.4% |
- −58% CTR at position 1 — Ahrefs, 300,000 keywords, Dec 2023 vs Dec 2025 (up from 34.5% in April 2025).
- Informational queries −61% (CTR 1.76% → 0.61%) — Seer Interactive, 3,119 queries, 42 organizations, 25.1M impressions.
- AI Overviews are spreading to money queries — informational share fell 91.3% → 57.1% while commercial rose to 18.6% and transactional to 13.9% (Semrush, 10M+ keywords, 2025).
Keeping it honest: AI Overviews don't automatically push everyone to zero clicks — tracking the same ~200k keywords before and after they gained an AI Overview, zero-click rate actually dipped slightly (33.75% → 31.53%, Semrush). The damage shows up as lower CTR on the clicks that remain. Medium
2. If people get answers from ChatGPT, Perplexity or Google AI instead of clicking, how do I still get seen?
By being cited inside the answer. Brands cited in a Google AI Overview earned about 35% more organic and 91% more paid clicks than uncited brands (Seer). Citation is the new visibility — though the evidence is correlational, and its link to actual leads is still unproven.
- Cited brands earned +35% organic clicks (0.70% vs 0.52% CTR) and +91% paid clicks (7.89% vs 4.14%) versus uncited brands — Seer Interactive, 3,119 terms, 25.1M organic + 1.1M paid impressions.
- Seer's own caveat: this is correlational — stronger brands may simply be more citable and higher-CTR. Read it as "queries where you're cited outperform queries where you're not".
No verified source connects AI-answer citation to actual leads or pipeline for B2B SaaS — only to clicks. Anyone claiming "GEO drives revenue" is ahead of the evidence. This is a real gap, and a good candidate for original attribution research.
3. What actually makes an AI engine cite my page?
Enrich the content itself. The peer-reviewed GEO study found adding quotations, statistics and on-page source citations lifts AI visibility by up to ~40% — while keyword stuffing does nothing. Freshness, semantic HTML and structured data also correlate strongly with being cited.
Tier 1 — peer-reviewed, experimental (GEO paper, Aggarwal et al., KDD 2024; ~10,000 queries, validated on Perplexity):
| Tactic | Visibility gain |
|---|---|
| Add quotations | ~41% |
| Add statistics | ~31–33% (often mis-cited as ~40%) |
| Cite sources on-page | ~30% |
| Keyword stuffing | ~0% — no effect |
"Up to 40%" is the best case for the method suite (37% on Perplexity), not an average, and it was measured on 2023–2024-era engines. A September 2025 B2B-SaaS preprint (GEO-16) adds the strongest on-page correlates of being cited: metadata & freshness (r=0.68), semantic HTML (0.65), structured data (0.63), on-page evidence & citations (0.61), authority (0.59). Medium — preprint, correlational
Which engine you target changes the game — citations available per answer, and what each favours:
| Engine | Citations / answer | Favours |
|---|---|---|
| Perplexity | 21.87 | Community (Reddit) |
| Google AI Overviews | 17.93 | Balanced pro + social |
| Gemini | 17.11 | — |
| ChatGPT | 7.92 | Authoritative KBs (Wikipedia) |
| Copilot | 2.47 | — |
Profound (680M citations, 2024–25) + Qwairy (669k citations, 8 providers, Q3 2025). Per-query inclusion is ~2.76× easier in Perplexity than ChatGPT. Source-mix percentages are volatile month to month.
4. Will Google penalize my AI-written content?
No — not for using AI. Google’s own docs say appropriate use of AI is not against its guidelines and reward content however it is produced. What gets penalized is scaled content abuse: mass-producing low-value pages to manipulate rankings.
| Practice | Google's actual stance |
|---|---|
| AI-assisted drafting of genuinely useful content | Allowed — ranks on quality |
| AI to research / structure original content | Explicitly endorsed |
| Mass-generating pages to manipulate rankings | Penalized (scaled content abuse) |
| Low-value output "however produced" | Penalized — the intent, not the tool |
Google Search Central (updated Dec 2025): using AI "to generate many pages without adding value for users may violate" the scaled-content-abuse policy, but AI "can be particularly useful when researching a topic, and to add structure to original content." In practice, ~86.5% of top-ranking pages already use some AI assistance and still rank (Ahrefs, 600k pages). The trigger is manipulative intent plus low value at scale — not the AI tool.
5. Is GEO a genuinely new discipline versus SEO — and is llms.txt worth doing?
GEO is real but built on top of SEO, not instead of it. 90–94% of AI Overview citations also rank in the organic top-20, and citation odds rise with rank. llms.txt, by contrast, has no proven payoff yet.
- 90–94% of AI Overview citations also rank in the organic top-20 (90% even for top-10) — seoClarity, 362,000 US keywords, 5.1M citations, Oct 2025.
- Citation odds scale with rank: position 1 is cited 43% of the time, position 3 31%, position 20 just 7%.
- But 44% of citations come from beyond the organic top-20 — ranking is necessary-but-not-sufficient, leaving real room for GEO-specific tactics.
So GEO is a distinct layer on top of SEO, not a replacement: keyword stuffing does nothing for AI answers, while quotations, statistics and on-page citations measurably help.
Is llms.txt worth it? No verified evidence that implementing llms.txt or granting AI-crawler access changes citations or traffic. Google has publicly called it "purely speculative for now." Our take: don't over-invest yet — it's cheap to add, but treat any promised payoff as unproven.
Contradictions & caveats
- Correlation ≠ causation for nearly every headline number — no controlled field experiment survived verification.
- Volatility: citation-source mixes swing hard (Reddit's ChatGPT share reportedly moved ~60% → ~10% within weeks). Several figures are single-day snapshots.
- Figure correction: the GEO paper's "add statistics" gain is ~31–33%, not the ~40% often quoted; ~40% is the best case for the whole method suite.
- Weakest link: the GEO-16 study is a non-peer-reviewed preprint that excludes ChatGPT — treat as indicative.
- Vendor incentive: the analytics vendors cited sell related tools; we relied on cross-vendor agreement on direction, not any single number.
Sources
Peer-reviewed / academic
- GEO: Generative Engine Optimization — Aggarwal et al., KDD 2024
- GEO-16 on-page citation correlates — Kumar & Palkhouski, Sept 2025 (preprint)
Platform documentation (first-party)
- Google Search Central — Using generative AI content (updated Dec 2025)
Original-data vendor studies
- Ahrefs — AI Overviews reduce clicks 58% (Dec 2025)
- Semrush — AI Overviews study (2025)
- seoClarity — AI Overview / organic overlap (Oct 2025)
- Profound — AI platform citation patterns (680M citations)
- Qwairy — provider citation behaviour Q3 2025
Secondary (corroboration)