# Google AI Mode: what it changes and how aegiszero.io is built for it

*Research brief — AegisZero (aegiszero.io), 2026-11-19. All sources listed at the bottom were opened during research.*

## TL;DR

1. Google AI Mode and AI Overviews retrieve content with a **query fan-out**: one user question is exploded into many sub-searches against Google's normal index, and the answers are synthesized. Classic crawlability and ranking remain the foundation — there is no separate "AI index" to game.
2. Citation data shows **classic ranking correlates but is not required** (studies disagree on how much comes from top-10: Ahrefs 38% vs older data 76% — a large minority of citations come from beyond page one).
3. **YouTube is the most-cited domain in AI Overviews** (~23% across industries per Surfer's 2026 report; Ahrefs tracks it as the fastest-growing too). A faceless YouTube-first content plan is therefore not a guess — it is the statistically dominant citation surface.
4. **Answer placement matters**: ~55% of AI Overview citations come from the top 30% of a page (CXL, 2026). Front-load answers; make sections self-contained.
5. **llms.txt is not a lever**: Google's search team states no AI system uses it and does not endorse it. Ship it as a courtesy, noindex it, and spend the energy on retrievable content instead.
6. **Measurement exists**: Search Console's generative-AI performance reports (launched June 2026) show impressions/citations in AI features — set the baseline on day one.

## How AI Mode actually works

Google's own documentation (Search Central, "AI features and your website") describes the mechanism: both AI Overviews and AI Mode may use a **query fan-out technique — issuing multiple related searches across subtopics and data sources** to build a response. Practical consequences:

- Your content must answer *sub-questions*, not just the head query. A page about "prop firm drawdown rules" gets decomposed into sub-queries like "what is a trailing drawdown", "when does a trailing threshold lock", "EOD vs intraday drawdown" — each one must find a self-contained passage to retrieve.
- Retrieval still happens against **Google's web index**, so indexing health, internal linking and clean HTML are prerequisites, not extras.
- The process shifts from query-focused to context-focused (Aleyda Solís's framing): the model reads passages in the context of the whole conversation, so ambiguity and buried answers kill retrievability.

## What the citation data says (2026 studies)

| Finding | Number | Source |
|---|---|---|
| Citations from the top 30% of a page | ~55% | CXL, Mar 2026 |
| Most-cited domain in AI Overviews | YouTube ~23.3%, then Wikipedia ~18.4%, Google.com ~16.4% | Surfer AI Citation Report, Sep 2026 |
| Citations from top-10 organic | 38% (Ahrefs) vs up to 76% in other datasets — studies diverge | Ahrefs Brand Radar / 15-study comparison, Nov 2026 |
| Beyond page one | Large minority of citations — page-one is not a gate | multiple studies |

Convergent takeaway: **rank well AND structure for passage-level retrieval** — neither alone is sufficient.

## Build actions implemented on aegiszero.io (v1)

1. **Question-shaped, self-contained chunks** — FAQ blocks, per-criterion methodology sections, "what does in-review mean" — each passage can be retrieved standalone by a fan-out sub-query.
2. **Answer-first sections** — key answers appear in the first lines of every section (the top-30%-of-page rule).
3. **Entity-consistent schema.org graph** — one `Organization` node (`#org`) with stable `@id`, reused by `WebSite`, `FAQPage`, `SoftwareApplication`, `Article` on every page. Entities disambiguate sub-query matching.
4. **One topic per URL, stable slugs** — `/methodology.html`, `/deals.html` etc., canonicals set; no client-side-only content rendering (all text is in the HTML source).
5. **Open crawl policy** — `robots.txt` allows all AI crawlers explicitly; no paywall, no JS-gated content.
6. **llms.txt shipped but noindexed** (`_headers` → `X-Robots-Tag: noindex`) following Google's own suggestion — zero-cost courtesy for on-page agents, no SEO expectation.
7. **Original research as content moat** — the risk-index methodology and the deals tracker produce facts that don't exist elsewhere, which is the only durable citation magnet.
8. **Measurement from day one** — Search Console property + generative-AI performance report as the AI-visibility baseline, plus the `/go/[slug]` click tracker for conversion truth.

## What we deliberately do NOT do

- No llms.txt hype or GEO snake-oil claims on the site itself — we cite Google's actual position.
- No templated AI-content spam (YouTube's 2025 inauthentic-content policy demonetizes it, and it contradicts the research-desk brand).
- No blocking of `Google-Extended`: AI Mode grounds in Google's index, so blocking Gemini-related crawlers would trade future surface for nothing we need.

## Measurement plan

- **Baseline (day 0):** Search Console property verified, sitemap submitted, gen-AI performance report bookmarked; screenshot first impression counts.
- **Weekly:** GSC AI-feature impressions; `/go/[slug]` click counts per firm; digest open rate.
- **Monthly:** manually probe AI Mode with 10 target questions ("which prop firm has the strictest trailing drawdown?", "is [firm] safe to pay?") and log whether/what gets cited. Track citation gain as the north-star AI metric.

## Sources (opened 2026-11-19)

- Google Search Central — AI features and your website: https://developers.google.com/search/docs/appearance/ai-features
- CXL — Where Google AI Overviews Cite From: A 100-Page Study: https://cxl.com/blog/google-ai-overview-citation-sources
- Surfer — AI Citation Report: https://surfer.ai/blog/ai-citation-report
- Digital Applied — AI Overview Citations: 15 studies compared: https://www.digitalapplied.com/blog/ai-overview-citation-studies-compared
- Search Engine Roundtable — Google Search Team Does Not Endorse LLMs.txt Files: https://www.seroundtable.com/google-does-not-endorse-llms-txt-40789.html
- Google Search Central blog — Search Generative AI performance reports (June 2026): https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
- Google support — Generative AI performance report: https://support.google.com/webmasters/answer/16984139
- Google blog — New opportunities, control and insights for website owners: https://blog.google/products-and-platforms/products/search/new-controls-website-owners
- CMSwire — Search Console gets AI visibility reports: https://www.cmswire.com/digital-experience/google-adds-ai-visibility-reports-to-search-console
- Conductor — Understanding Query Fan-Out: https://www.conductor.com/academy/query-fan-out
- Surfer — Query Fan-Out: https://surfer.ai/blog/query-fan-out

## Application for webxon (client-facing framing)

This research maps directly onto a sellable service: **AI-visibility audits** (does a client's content survive fan-out sub-queries?), **schema/entity implementation**, **answer-first content restructures**, and **GSC gen-AI reporting setup**. The aegiszero.io build is the reference case study: a small site engineered for AI retrieval at zero infrastructure cost, with measurement in place from day one.
