← Field Notes
Strategy11 August 2026·10 min read·Chris Ma

THE
THREE
GATES.

How answer engines decide what to cite, and why most content fails before you've written a word

AEOGEOSEOContent StrategyBrightEdge

Most content optimisation assumes a visibility problem. It usually isn't. It's a mechanism problem. The engine never opened your page, or chose not to click through, or couldn't extract a usable answer from what it found. Three different problems. Three different fixes.

Treating them as one is why most AEO advice gives you motion without traction.

Key Takeaways
  • Content failing Gate 2 (Chosen) is the most common AEO problem: the engine never clicked through because nothing in the snippet signalled a usable answer.
  • BrightEdge data shows AI engines are opening a progressively smaller share of candidate results per query, making selection at Gate 2 increasingly competitive.
  • The three-gate diagnostic maps to three distinct fixes: technical (Gate 1, Fetchable), snippet and title optimisation (Gate 2, Chosen), and content structure (Gate 3, Extractable).
  • Traditional SEO optimises for ranking. AEO optimises for being opened, extracted from, and cited. These are three different mechanisms requiring three different interventions.
01

THE CORE REFRAME

#
The rule

The model can only answer from what it fetched. Not what exists. Not what is true. What it actually pulled into context for that specific query.

This sounds obvious stated plainly. It's not obvious in practice, because most content optimisation still treats AI visibility the same way it treats search ranking: quality, relevance, and authority. Those things matter. But they're evaluated only after the engine has already decided to open your page. Get eliminated before that decision and quality is irrelevant.

The mechanism works in sequence. A human types a messy, multi-part question. The engine doesn't search for that exact query — it translates it into several clean, query-shaped searches (query fan-out). Each search returns its own candidate source list. The engine then decides which candidates to open, which pages to extract from, and finally which extracted chunks to use in composing the answer. Three decisions. Three places to fail.

USER PROMPT"best mattress for backpain and hot sleepers"QUERYFAN-OUTbest mattress for back pain and hot sleeperscooling mattress comparison 2026mattress for lumbar support reviewshybrid vs memory foam back supporteach query returns its own candidate source list; engine decides which to open
Query fan-out — one human prompt becomes multiple discrete searches, each with its own candidate source list.
02

THE COVER

#

Before any extraction, before any quality assessment, the engine makes a click decision based on what's visible without opening the page: URL, title, snippet, sometimes a freshness date. That's the cover. Content quality inside the page is irrelevant at this stage. If your title signals brand story rather than utility, if your snippet reads like a press release rather than an answer, the engine won't click through regardless of what's inside.

This is why a well-written product page with strong SEO fundamentals can still be invisible in AI answers. The cover wasn't shaped for the engine's selection criteria. Optimizing the inside without optimizing the cover is backwards.

03

THE THREE GATES

#

Use this diagnostic any time visibility is lower than expected. It tells you which problem you have rather than asking you to guess.

GATE 01: FETCHABLECan the engine reach and ingest the page?Indexed?robots.txt clear?No paywall / auth?Low visibility scoreGATE 02: CHOSENDoes the cover signal the right answer?Title utility > brand?Snippet answer-shaped?Format matches intent?Good score, low rank (9th, 10th)GATE 03: EXTRACTABLECan the engine lift a clean chunk?Not buried in accordions?Text-based, not image/video?No JS render-blocking?Cited but content missing from answerCITEDINANSWERfailing any gate removes you from consideration. diagnosis tells you which one
Three gates — failing any one removes you from consideration for that query.
01FETCHABLE

Can the engine reach and ingest your page?

  • Indexed and not blocked by robots.txt?
  • Not behind a paywall or authentication?
  • Crawl errors in GSC?
SymptomLow visibility score across platforms — you're not in the candidate list at all.
02CHOSEN

Does your cover look like the answer?

  • Title signals utility, not brand?
  • Snippet is answer-shaped, not promotional?
  • Format matches the query intent — comparison vs single review?
SymptomDecent visibility score but low rank (9th, 10th position). You're in the list but not being selected.
03EXTRACTABLE

Can the engine lift a clean, usable chunk?

  • Answers visible in plain HTML, not JavaScript-rendered?
  • Not buried in accordions or tabs?
  • Text-based — not locked in images, charts, or video with no surrounding copy?
SymptomBeing cited but content isn't appearing in the actual answer. The engine found you but couldn't pull a clean extract.
Platform divergence is diagnostic

Ranking first on ChatGPT and tenth on Gemini isn't noise: it means two engines are applying different selection logic to the same content. The fix for each will be different. Averaging them into a single "AI visibility" number hides both the problem and the opportunity. Diagnose each platform separately.

04

SCORE VS RANK

#

These two metrics get conflated constantly. They measure completely different things.

Visibility Score

Your raw presence in AI answers. This can rise for everyone simultaneously as engines cite more sources overall. A rising score tells you the category is growing.

Visibility Rank

Your position relative to competitors. This is the number that tells you whether you're actually winning. Track both, but optimise for rank.

A rising score with flat or falling rank means the category is growing but you're not capturing more of it. This is the most common misread in AEO reporting — a chart that goes up and right masking a competitive position that's quietly getting worse.

05

THE SAGE LOOP

#

SAGE is a four-stage cycle (Setup, Analyse, Generate, Engineer) and its most useful property is that it tells you which stage to be in right now. The temptation is to default to Generate (producing content) because it feels productive. Generating against the wrong gaps, or without measurement, is how you ship a lot and move nothing.

SETUPDecide what to trackANALYSEFind where you're losingENGINEERMake wins repeatableGENERATETurn gaps into shipped workLOOP"What time is it?" / which stage you're in, right nownot a checklist you complete once; a cycle you run continuously
SAGE loop — a cycle you run continuously, not a checklist you complete once.
SETUP"We don't know what to track yet"
  • Define 5–10 prompts using real buyer language, not target keywords
  • Validate each has genuine query volume
  • Map 3–5 direct competitors for citation comparison
  • Tag by funnel stage and persona for later filtering
ANALYZE"We're not sure what's missing or why we're losing"
  • Run the three-gate diagnostic on 5 lowest-visibility priority pages
  • Pull citation data — see which competitor sources are winning compound-job queries
  • Check platform divergence — diagnose each engine separately
  • Identify 3 "should be winning but isn't" pages
GENERATE"We know the gap but nothing is shipping"
  • Prioritize highest-leverage fixes first — not the easiest ones
  • Rewrite one existing page as a utility asset as a test before building net-new
  • Re-run visibility check 2–4 weeks post-publish — never assume it worked
ENGINEER"It's working but it's manual every time"
  • Set recurring monthly minimum SAGE review
  • Build automated alerts for visibility drops on priority pages
  • Write a one-page reusable brief template so wins aren't personality-dependent
06

CONTENT THAT GETS CITED

#

Two concepts do most of the work here.

Compound jobs — queries that combine multiple needs in a single ask ("cooling and support", not just "cooling"). These force the engine to be selective. It can't satisfy every angle, so it picks the source that covers the combination best. Single-angle content loses to content that resolves the compound job in one place. This is where differentiation concentrates, because most content is still written to rank on one keyword at a time.

Utility assets — content shaped to match what the engine is already fetching: comparative, high-density, answer-shaped. Explicitly not: product pages, brand-story copy, buried leads. The format question to ask before writing is "does this match the shape of the answer the AI is already producing for this query?" — not "does this match our content brief?"

UTILITY ASSET▶ What it is + who it's for▶ Comparison: X vs Y vs Z▶ Answer: cooling AND support▶ Verdict per use case▶ Direct recommendationresolves the compound job in one placePRODUCT PAGE◦ Brand story / heritage◦ Feature list◦ Social proof section◦ Single-angle positioning◦ CTA / request demosingle angle. loses compound-job queriesVS
Utility asset vs product page — same topic, different shape, very different citation rate.
Supply chain awareness

Every AI-generated answer has a traceable supply chain: specific sources, specific citations, specific reasons one brand gets mentioned over another. Most marketers never read this chain. Understanding which sources an engine draws from — and why those rather than others — shows you what content shape actually wins for a given query before you've written a word.

07

APPLYING THIS TO THE STACK

#

The three-gate model maps directly onto existing tools. The diagnostic is faster when you already know which data to pull.

BrightEdgeGate 2 → 3

Already tracks some AI-answer visibility — cross-reference its citation data against the three-gate model rather than treating an AI visibility drop as one undifferentiated problem. The citation data tells you which sources are winning; the gate diagnostic tells you why.

GSCGate 1

Crawl errors and index-coverage data map directly onto Gate 1 (Fetchable). Check this first, before assuming a content or positioning problem. A crawl block or indexing gap is a one-step fix; a content positioning problem is a quarter of work.

AEMGate 3

Audit page templates against the extraction principle. Are answers buried in accordions or tabs — a common AEM pattern — that block Gate 3? Extractability is a template-level problem, not a content-level one. Fixing the template fixes it across every page using it.

MarketoGate 2

Campaign landing pages built brand-first rather than utility-first are classic Gate 2 failures. A page optimized for a compound job (X vs Y for [use case]) will outperform a single-angle product page in AI citation terms even with identical SEO fundamentals. Rewrite one as a test before rebuilding the whole set.

The SAGE framework, three-gate model, query fan-out, and utility asset framing are sourced from Profound University's Profound 101 curriculum — Profound is the AEO analytics platform named sole Leader on G2's first AEO Grid (Winter 2026). The underlying framework is platform-agnostic. Product-specific performance claims from their marketing material should be treated with more scepticism than the model itself.

Recommended Reading

Marcus Sheridan · Wiley

The method that anticipated answer-engine optimisation before the term existed — full transparency, direct answers, and genuine expertise as the content strategy.

Google Search Central · developers.google.com

Google's own explanation of crawling, indexing, and ranking — reading primary source documentation removes the guesswork from optimisation decisions.

Andy Crestodina · Orbit Media Studios

A data-driven content marketing handbook from someone who has surveyed thousands of bloggers annually for a decade — grounded in evidence rather than theory.

Continue the conversation

If this changed how you think about it — or you think I'm wrong — I want to know.

Corrections, disagreements, and applications all welcome. Replies go directly to Chris.

Get in touch →
Field Notes · PodcastHost + Expert · Gemini TTS

THE THREE GATES

~6-8 min

1× · Two speakers · tap to play