Key takeaways
- For dev tools, documentation is the main asset AI systems retrieve. Blog posts come second and landing pages a distant third in how models answer technical questions.
- Software prompts in ChatGPT lean on GitHub, vendor docs and review sites. Your own site, docs and a public GitHub presence are your main citation sources.
- Coding agents like Claude Code and Codex fetch docs differently from consumer AI search. Treat docs as a separate channel and keep them crawlable, accurate and available in clean formats.
- You need three layers: a measurement tool for AI visibility, a content workflow with engineers on the byline, and technical hygiene so crawlers can read your site.
- An agency can help, but pick on who actually writes the content and how fast they ship, not on client logos.
Why dev tool SEO is its own problem
SaaS SEO playbooks assume a buyer who reads comparison pages and pricing pages. Developers don't behave that way. They search for "how to implement OAuth2 in Node.js" or "PostgreSQL vs MySQL for time-series data", and increasingly they skip search entirely and ask a model.
That changes what you optimize. Pepper Content's 2026 guide to GEO agencies for developer tools argues that docs, tutorials and reference pages are what models retrieve, and that accuracy is binary: a deprecated API call or an outdated code sample does more damage than publishing nothing. I agree with that. A stale snippet burns trust with the reader and with the model that learned from it.
The touchpoints are also broader than a website. Developers form opinions in documentation, API references, GitHub repositories, migration guides, tutorials, Google, ChatGPT, Perplexity and AI Overviews, long before talking to sales. Hackmamba's devtools SEO guide makes the same point: the work combines technical content, documentation, topical authority, technical SEO, digital PR and generative engine optimization.
What AI systems actually cite for software prompts
Before picking tools, look at the evidence on what gets cited. Promptwatch Data's report on software prompts in ChatGPT (Sep 28 to Oct 4, 2026, organic prompts only) shows the pattern:
- GitHub is the most cited software domain at 1.2% of software citations.
- learn.microsoft.com follows at 0.98% and G2 at 0.73%.
- AWS docs get 0.41%, Google Cloud docs 0.35%, Gartner 0.34% and Capterra 0.32%.
- Docs and help centers together earn 2.64%, vendor sites 2.01%, review sites and analysts 1.57%, and app stores 0.47%.
Vendor sites plus docs earn 4.65%, roughly three times what review sites and analysts get. The practical reading: your own domain is your main citation source, and you can't outsource it to G2 listings. Still keep G2 and Capterra profiles complete, because they do show up.
Which content types win
Promptwatch's ChatGPT citation share by content type report (same week) gives the broader picture across all prompts:
| Content type | Share of ChatGPT citations |
|---|---|
| Product pages | 22.4% |
| Landing pages | 16.8% |
| Listicles | 13.6% |
| How-tos | 8.7% |
| Documentation | 7.5% |
| General articles | 6.1% |
| Comparisons | 5.9% |
| News | 5.7% |
| FAQ pages | 3.3% |
| Reviews | 1.7% |
For informational prompts, which is where most developer questions live, documentation rises to 8.5% and how-tos to 10.7%. For commercial prompts, product pages, landing pages, listicles and comparisons make up almost 70%. So a dev tool needs both: strong docs and how-tos for "how do I" questions, and sharp product and comparison pages for "which tool" questions.
Freshness matters too. Promptwatch's content half-life report found that landing pages and docs stay cited the longest, while news loses half its pages within 36 days. Changelog-style announcements fade fast. Evergreen docs don't.
Docs are a separate channel: coding agents
This is the part most generic AI SEO advice misses. Coding agents don't behave like ChatGPT Search.
Promptwatch's markdown research found HTML makes up 99.94% of AI search citations and markdown files just 0.05%, roughly 2,000 to 1. So don't build .md mirrors of your blog or marketing pages hoping for citations. But the same logs show a big uptick in docs crawling by Claude's bot (powering Claude Code) and OpenAI's search bot (powering Codex), with top logged paths being /docs/.../*.md endpoints, the busiest at 3,241 requests. Agent-native search engines like Exa, integrated into tools such as OpenCode, let agents search and fetch URLs on their own.
The takeaway is simple:
- Keep marketing and blog content as clean, server-rendered HTML with a clear heading hierarchy and direct answers.
- Fix heavy client-side JavaScript rendering first. If a crawler can't see your docs, nothing else matters.
- Reserve .md endpoints (and an llms.txt) for developer docs, where agents fetch them directly.
- Version your docs clearly so agents don't recommend a deprecated endpoint.
Crawler traffic is also shifting quickly. Promptwatch tracked Meta-WebIndexer going from about 2% to nearly 38% of AI crawler requests between mid-July and August 9, 2026, and publishes weekly verified AI crawler traffic. Check your robots.txt and CDN rules against the current list of bots, not last year's.
Community channels are less reliable than they were
Reddit and community threads used to be a dependable dev tool channel. In ChatGPT, that has changed. reddit.com's share of ChatGPT citations fell from about 4% to 0.5% on August 14, 2026, and from Sep 28 to Oct 4 ChatGPT got 0.46% of citations from social platforms while Google AI Overviews got 12.2%. So community presence still matters for Google AI Overviews, but don't build a ChatGPT strategy on it. Put the effort into your own docs and content first.
The tool stack
You need tools for four jobs: measuring AI visibility, seeing crawler behavior, producing content, and classic SEO hygiene. Here is how I would split them.
| Job | What to look for | Example tools |
|---|---|---|
| AI visibility tracking and optimization | Real-UI tracking, citations, crawler logs, action lists | Promptwatch, Profound, Peec AI, Otterly.AI |
| Traditional SEO suite with AI add-ons | Keyword and backlink data plus brand tracking | Semrush, Ahrefs Brand Radar, SE Ranking |
| Enterprise technical SEO | Large site crawling, log analysis | Botify, BrightEdge, Conductor |
| Content optimization | Briefs, scoring, refresh | Surfer SEO, Clearscope, MarketMuse, Frase |
| Programmatic and scraping infrastructure | Data for integration and comparison pages | Firecrawl, Apify, Bright Data |
AI visibility and optimization
Most devtools teams start by asking "are we named when someone asks for the best API for X?" Plenty of tools answer that. The harder questions are why you aren't, and what to change.
Promptwatch is the one I would put first for dev tool companies, partly because it tracks coding agents (Codex, Claude Code and OpenCode) alongside ChatGPT, Perplexity, Gemini and AI Overviews, and partly because its IP-verified crawler logs show whether bots are actually reading your docs and hitting errors. It then connects that to citations, AI-driven visits, and an action list, and its Content Agent can draft and publish into Webflow, Framer or WordPress. Pricing starts with a free tier and paid plans from $95/month; confirm current plans at promptwatch.com/pricing.
Other trackers are worth a look if you want a narrower scope. Profound is an enterprise-grade option with strong monitoring.
Peec AI and Otterly.AI are lighter prompt trackers that suit teams that only need to watch mentions.
If you want a broader comparison, the GEO software directory at bestgeosoftware.com lists many more options.
Suites you may already pay for
If you already use Semrush or Ahrefs, their AI features are a cheap way to start. They work off fixed prompt sets and don't give you crawler logs, so treat them as a first look rather than a full answer.
Enterprise technical SEO
Large API platforms with thousands of reference pages often need serious crawling and log analysis. Botify is built for that kind of site.
Content optimization and data collection
For refreshing how-tos and comparison pages, Surfer SEO and Clearscope score drafts against search data. Neither replaces an engineer checking code samples.
For programmatic pages built from integration catalogs or public specs, a crawling API like Firecrawl gives you clean, LLM-ready content.
A practical playbook
1. Audit what AI can read
Check robots.txt for AI bots, confirm your docs render without JavaScript, and look at crawler logs to see which pages bots hit and which return errors. Google's May 2026 note on optimizing for generative AI in Search stresses valuable, unique, non-commodity content, which is a polite way of saying generic explainers won't earn citations.
2. Build a prompt set from real developer questions
Pull questions from support tickets, GitHub issues, Discord and Stack Overflow-style searches. Split them into informational ("how do I paginate results with X API"), comparison ("X vs Y for webhooks") and branded prompts. Track them across models, including the coding agents, since an answer inside Claude Code can differ from one in ChatGPT.
3. Fix docs before writing new posts
Publish working code samples, version numbers, error-code references and migration guides. Make sure each page answers one question directly near the top. Remove or redirect deprecated pages.
4. Write comparison and how-to content with engineers on the byline
Comparisons and how-tos earn citations, but only if they are accurate. Have a practicing engineer review every code block. A content tool can draft; a human should verify.
5. Keep GitHub, G2 and your product pages current
GitHub was the top cited software domain, so README quality, topics and descriptions matter. Update review profiles and make sure product pages state supported languages, SDKs and limits plainly.
6. Measure traffic, not only mentions
A mention that never sends a visit is a vanity metric. Track visits from AI platforms, signups attributed to them, and which pages drive both. Refresh pages whose citations decay.
Should you hire an agency?
Sometimes. If your team has no one who can write technical content, an agency saves months. Compare candidates on who writes, not on logos.
| Agency | Angle | Notes |
|---|---|---|
| Hackmamba | Long-term technical content, SEO and AI search | Named by its own guide as best for a growth partner |
| Draft.dev | Devtools only, 300+ engineer-writers | From $9,000/mo, 3-month minimum per Pepper |
| EveryDeveloper | Developer experience audit plus SEO content strategy | Custom pricing |
| Infrasity, Onely, Megawatt | Varied: technical SEO, content, enterprise work | Quote-based |
Note that most of these lists are published by agencies that rank themselves, so read them as starting points. Maintouch's own list argues execution speed matters more than strategic polish for startups whose engineers are busy building features, which is a fair filter: ask how much of the work lands on your team.
If you want a general SEO and GEO partner rather than a devtools specialist, 1001 SEO Media is a full-service agency covering technical SEO, content and AI search work, with no long-term contracts. Ask any agency, including us, to show a sample piece written by an engineer and to explain how they will measure AI citations.
Common mistakes
- Optimizing the blog while docs sit behind client-side rendering.
- Creating markdown mirrors of marketing pages. The citation data says it doesn't pay.
- Shipping AI-generated tutorials without running the code.
- Chasing Reddit for ChatGPT visibility after the August 2026 drop.
- Treating a visibility score as the goal instead of signups and usage.
Where to start this week
Run a free visibility scan, list 30 real developer questions, and check which of your docs pages bots actually fetch. Then fix the three biggest crawl or accuracy problems before spending a dollar on new content. For more rank tracking options, the directory at ai-rank-tools.com is a reasonable next stop.