In the last couple of months I started building projects again after a long sabbatical. After an initial research on what changed since the last time I built a commercial product, an open-source library or a software project in general, it became clear that discoverability has become a critical component in a successful product lifecycle - now more than ever: which is how I started specializing in AI Visibility.
This is the first post in a series based on the awesome-ai-visibility repository: a concise, open-source playbook for making your docs, APIs, and projects visible to AI systems. This post is the map; later posts go deep on each layer.
What is AI visibility?
AI visibility is the degree to which AI systems - ChatGPT, Claude, Gemini, Perplexity, and AI coding assistants - can discover, understand, cite, and recommend your website, documentation, APIs, or open-source project. For developers, it spans crawlable docs, machine-readable specs, retrieval-friendly content structure, consistent entity signals, and measurable mentions across models.
You'll see overlapping terms for parts of this problem:
| Term | What it optimizes | Core tactic |
|---|---|---|
| AEO (Answer Engine Optimization) | Direct answers in AI assistants and AI Overviews | Question-shaped headings, self-contained answers |
| GEO (Generative Engine Optimization) | Being cited inside generated responses | Citations, statistics, quotable claims |
| AI SEO / LLMO | Umbrella: overall presence in LLM outputs | Crawlability + structure + entity consistency |
These aren't competing disciplines. For developers they converge on one artifact: your documentation. And the tactics are measurable — the original GEO research paper (Princeton/IIT Delhi, KDD 2024) found that adding citations, quotations, and statistics improved a source's visibility in generative engine responses by up to ~40%.
Why this matters now: AI is the new distribution channel
Two mechanisms decide whether a model recommends you:
- Training data: your project's footprint (docs, GitHub, tutorials, discussions) at training time. This is slow-moving, compounding authority.
- Retrieval (RAG): what AI search and coding assistants fetch live from the web. This is where you can win this quarter, if your content is crawlable and citable.
AI-referred traffic is still small in volume, but it converts dramatically better — 14.2% vs. 2.8% for classic organic search in one benchmark — because the assistant has already qualified the recommendation. For a dev tool, one citation inside a Cursor or ChatGPT answer is worth more than a page-two Google ranking.
The AI Visibility Playbook (condensed)
![The AI visibility stack for developers: crawlable foundation, machine-readable docs, answer-ready content, and measurement]
The full checklist lives in the awesome-ai-visibility repo. Here is the condensed, technically honest version:
-
Publish
llms.txt— a Markdown file at/llms.txtmapping to canonical docs, API references, examples, and changelogs (spec by Answer.AI, 2024). Treat it as an emerging convention: cheap to add, but not a replacement for crawlability or SEO fundamentals — major engines haven't confirmed they consume it. - Ship machine-readable docs and specs — accessible HTML/Markdown with stable URLs, plus OpenAPI/AsyncAPI/GraphQL schemas and versioned SDK references where applicable.
-
Structure content for retrieval — descriptive
H1–H3headings, one question or task per section, and sections that stay understandable when retrieved independently (that's how RAG chunks them). Keep code, prerequisites, and expected output together; split at semantic boundaries, not arbitrary token limits. - Write answer-ready content — answer "what is it, when should I use it, what are the trade-offs" directly, with FAQs, troubleshooting, and runnable examples.
-
Fix technical discoverability —
robots.txtthat doesn't accidentally blockGPTBot,ClaudeBot, orPerplexityBot; XML sitemaps; canonical URLs; and server-rendered key content (most AI crawlers execute little or no JavaScript). - Keep entity signals consistent — same product/org name, URL, and description across your repo, package registry, docs, and socials; add JSON-LD structured data where it disambiguates.
- Make claims reproducible — methodology, dates, and limitations turn your pages into citable sources. This is exactly what GEO research rewards; stale or vague content gets skipped.
- Measure and iterate — track prompts, mentions, citations, and accuracy across models over time; re-test after each release and fix weak source pages first.
Each of these is a checkbox in the repo, with curated tools and references — so you can treat AI visibility as an engineering task with a definition of done, not a marketing vibe.
FAQ
Is llms.txt worth adding in 2026?
Yes, because it costs minutes and documents your canonical URLs — but do it after the fundamentals. No major AI provider has confirmed using llms.txt for ranking or retrieval, so crawlable, well-structured docs still do the heavy lifting.
Does traditional SEO still matter for AI visibility?
Yes. AI search products ground their answers in web indexes and live crawls, so crawlability, canonical URLs, and clean information architecture remain the substrate AEO and GEO build on. SEO gets you retrieved; AI visibility work gets you cited and recommended.
How is AI visibility different from SEO?
SEO optimizes for ranked lists of links judged by humans. AI visibility optimizes for being retrieved, understood, and named inside a single synthesized answer - which rewards self-contained sections, verifiable claims, and consistent entity signals more than keywords.
How do I measure AI visibility?
Define a prompt set that matters for your project ("best X for Y"), run it across ChatGPT, Claude, Gemini, and Perplexity on a schedule, and track mentions, citations, and accuracy over time. The measurement section of awesome-ai-visibility lists tools that automate this.
What's next in this series
This post is the overview. Next up, one deep-dive per layer of the playbook: llms.txt and machine-readable docs, retrieval-friendly documentation structure, and measuring AI visibility across models.
If you find this useful, star and contribute to awesome-ai-visibility - it's a community-maintained list, and PRs with maintained, developer-relevant resources are welcome. Follow agentic-devrel for the rest of the series.
Top comments (1)
AI visibility is becoming a developer problem because the inputs are technical: crawlability, structured facts, clean product data, canonical pages, and answerable documentation. The marketing layer still matters, but if the site cannot be parsed and trusted by machines, the brand story never reaches the answer.