GEO and AEO for DevRel | DevRel Almanac

GEO covers attempts to increase a product’s inclusion and citation in generated answers. AEO covers structuring content so an answer engine can extract a direct response. The terms overlap in practice:

Generative Engine Optimisation (GEO). Optimising for citation and inclusion when generative AI systems (ChatGPT, Claude, Perplexity, Gemini) produce answers. The headline metric is Share of Model or Share of Voice: how often your brand is mentioned in responses to category-relevant prompts.
Answer Engine Optimisation (AEO). Structuring content so AI answer engines can extract clean, direct answers. Overlaps heavily with classical SEO best practices (Q&A formatting, schema markup) but tunes them for conversational extraction rather than ten-blue-links ranking.

Evidence for GEO and AEO varies. Some practices overlap with established documentation and SEO work; newer claims about AI citations need separate measurement.

Where AI assistants get their answers from

A practical breakdown:

Assistant	Primary index / retrieval source	Notes
ChatGPT	Bing (via OAI-SearchBot for live retrieval); training data	Bing Webmaster Tools verification matters; allow `OAI-SearchBot` separately from `GPTBot` (training-data crawler)
Perplexity	Live web retrieval (PerplexityBot)	Citations shown inline; detailed ranking criteria are not public
Gemini	Google Search index + Google’s broader retrieval	Strong SEO is prerequisite; YouTube videos with `VideoObject` schema often cited for “how-to” queries
Claude	Live web retrieval (ClaudeBot, Claude-User); training data	Anthropic has not published detailed retrieval mechanics
Copilot (Microsoft)	Bing; specific developer-tool data	DevTools-specific context

DevRel teams optimising for AI visibility need to satisfy multiple systems, each with slightly different mechanics. There is no single GEO playbook that works equally well across all of them.

Foundational GEO practices

These work across most assistants:

1. Authoritative, substantive content

Retrieval systems can inspect the following public evidence and provenance signals:

Original engineering analysis with enough detail to support its claims.
Citations from other authoritative sources.
Long-running domain history.
Wikipedia presence (if applicable).
Mentions in trusted publications.

Keyword-stuffed pages provide little evidence for a technical answer and may perform poorly in both search and retrieval.

2. Clear semantic structure

Headings as questions (“How do I X?”) or canonical topic titles.
One concept per page.
Lead with the answer in the opening paragraph.
Bullet-list and table formats where appropriate.

3. Schema.org / JSON-LD markup

For developer-product content particularly useful:

Article with dateModified, author, publisher.
HowTo for tutorials.
FAQPage for FAQ-style content.
SoftwareApplication for product pages.
Organization and Product schema for brand consistency.
VideoObject for embedded videos.

Schema exposes metadata in a defined structure rather than leaving a crawler to infer it from page layout.

4. Crawler permissions

Audit your robots.txt to ensure AI crawlers can read what you want them to read. See Agent-facing documentation for specifics. Many teams accidentally block AI crawlers while trying to block aggressive SEO scrapers.

5. Cross-platform consistency

Keep product descriptions consistent across the homepage, documentation, README, Wikipedia entry, and third-party listings. Contradictory descriptions can produce inconsistent answers or explicit uncertainty.

6. Freshness signals

Update dateModified when content changes, name versions explicitly, and mark or remove deprecated material. These provide live retrieval systems with explicit freshness and version signals.

Developer-product-specific tactics

Beyond general GEO, several practices that matter especially for DevRel:

Maintain a canonical “What is [product]?” page

Maintain a short, factual page that states what the product does, who it serves, and its important constraints. This gives retrieval systems a canonical passage separate from changing campaign copy.

Maintain “X vs Y” comparison content

Publish factual comparison content for the relevant alternatives in your category. These pages give assistants a first-party source for questions about differences and trade-offs.

State factual differences, trade-offs, and the cases where another product is a better fit. A promotional comparison that hides those trade-offs supplies incomplete evidence for the query.

Earn relevant third-party coverage

Get mentioned positively on:

Relevant Reddit communities, which some AI assistants weight heavily.
Hacker News (substantive Show HN launches and threads).
dev.to and Hashnode posts by independent developers.
YouTube tutorials by trusted educators.
Podcasts with transcripts.
High-authority newsletter mentions (Console.dev, TLDR, etc.).
Stack Overflow answers (the archive is still cited despite the platform’s decline).

Mentions in these channels provide independent sources that retrieval systems may use. Track whether such coverage is associated with changes in citations rather than assuming the relationship.

YouTube transcripts make video content searchable and available to retrieval systems. Publish a complete transcript when a tutorial contains information not available elsewhere.

The trade-off: video is expensive to produce. DevRel teams using video for GEO in 2026 treat it as an AI-visibility strategy and a brand format.

Publish in relevant external publications

External publication provides another attributable source and distribution channel. When cross-publishing, use canonical URLs on:

dev.to and Hashnode.
LinkedIn long-form posts and newsletters.
Substack or Beehiiv.
Medium (less weighted than 2018 to 2020 but still cited).
Industry publications (TheNewStack, InfoQ, occasional Smashing Magazine, etc.) where the editors will accept submissions.

Tools for monitoring brand presence in AI

A new tools category emerged in 2024 to 2026. Among the better-known products:

Profound (tryprofound.com). Measures brand visibility across major AI assistants; tracks Share of Model and competitive positioning.
AthenaHQ. AI-search analytics.
Otterly.AI. Similar tracking of AI-mention frequency.
Peec AI. Tracks brand mentions across AI engines.
LLMrefs. Citation and authority tracking.
Goodie tools. GEO-focused stack.
Sight AI (trysight.ai). AI search visibility tracking.
Pure Visibility. GEO platform.

The tools use different methodologies and can report substantially different numbers. Use a consistent prompt set for relative tracking over time rather than treating any result as an absolute measure.

Reasonable metrics for DevRel teams

Among the metrics worth tracking:

Share of Voice / Share of Model. Across a defined set of category-relevant prompts, what percentage of responses mention your brand?
Citation accuracy. When AI assistants describe your product, what fraction of technical claims they make are correct?
AI-mediated signup attribution. When you ask new signups how they heard about you, what share say “ChatGPT,” “Claude,” “Perplexity,” or “an AI assistant”?
Authority-domain presence. Are you mentioned on Reddit, HN, GitHub READMEs, prominent YouTube channels?
Citation variance. Measure how much results change across months, models, and prompt phrasings; stability alone does not establish authority.

Caveats

Two caveats:

AI citations are noisy. Multiple analyses through 2026 report 40 to 60% monthly variance in AI citations across the same prompts. AI assistant models update frequently; context windows shift; retrieval algorithms change. Treat your GEO position as a trend, not a fixed measurement.
AI assistants do not always cite their sources. ChatGPT often answers without showing links, so referral data cannot measure all AI-mediated discovery. Lack of a citation does not establish lack of influence.

What to deprioritise

These practices have weak evidence of value as of 2026:

Stuffing keywords into AI-readable surfaces. It adds no technical evidence for the answer.
Generating large volumes of AI-written content optimised for “AI search.” Without accountable authorship and review, it can introduce generic or inaccurate material.
Treating llms.txt as a complete GEO strategy. It’s one signal, not the strategy.
Optimising only for ChatGPT. Retrieval mechanics differ across assistants, so results from one do not necessarily generalise.

A 90-day GEO program for a DevRel team

Months 1: Baseline. Run prompts across all major assistants. Audit robots.txt. Run a schema.org audit. Verify Bing Webmaster Tools setup. Identify three to five high-value pages on your domain that need rewriting.

Month 2: Optimise. Rewrite the high-value pages with the practices above. Publish canonical “what is X?” and “X vs Y” pages. Coordinate one major YouTube collaboration with a trusted educator. Ensure all SDK READMEs have date / version / canonical URLs.

Month 3: Measure and iterate. Re-run the prompt set and compare the results with the baseline. Use the month-three results for the next scheduled review.