SquishPixel Blog · AI search · SEO · Web performance

A six-step process for getting cited in AI answers, from audit to monthly tracking

DEV Community published a step-by-step guide to generative engine optimization, or GEO. The guide, originally posted on RankVelt's blog, defines GEO as improving a brand's presence inside AI-written answers through citations, mentions and recommendations. It says GEO builds on SEO foundations such as crawlability, relevance, authority and content quality. The difference is the goal: being named or cited in an answer, whether or not anyone clicks.

The first step is a baseline audit. You list the questions customers ask AI tools, then test the same prompts in ChatGPT, Gemini and Perplexity, signed out or in private windows. For each prompt you record whether you are mentioned, cited, recommended or absent, and who is cited instead. The guide says answers vary by session and location, so trends matter more than single checks.

The second step is crawler access. The guide says to check robots.txt (the file that tells bots which pages they may visit), server and CDN rules, login walls, paywalls and JavaScript-only content. It also says Google has stated that llms.txt does not improve visibility in Google Search, and that no special AI schema is needed.

The remaining steps cover content and measurement. Pages should put a direct answer near the heading, back it with evidence and use question-style headings, lists and tables. Brands should keep facts consistent across their site and profiles, add Organization and LocalBusiness schema, and earn reviews and independent mentions. The guide suggests small per-engine adjustments, but not separate page versions. It ends with monthly re-testing of the same prompts, plus checks on AI referrals in analytics and Search Console AI data where available.

Why it matters

The guide treats AI visibility as an extension of ordinary SEO work, not a new discipline. If your pages cannot be crawled or do not answer questions clearly, the guide says there is nothing for AI systems to cite. That makes basic technical checks a sensible first move.

The guide comes from RankVelt, which sells GEO services, so treat it as one practitioner's view. Still, its warnings are practical: avoid hacks, avoid mass-producing thin pages for prompt variations, and do not judge results from a single test. Its advice to track a fixed set of prompts gives you a simple way to see change over time.

Source: DEV Community. This brief summarises their reporting; read the original for the full detail.