What Google actually published
Google’s generative AI optimization guidance for Search (AI Overviews, AI Mode) stresses useful, crawlable, authoritative pages — the same systems behind classic results. It explicitly says you do not need special AI-only files, artificial “chunking,” or inauthentic mention schemes to appear in those features.
On llms.txt: Google Search does not use it as a ranking or AI-visibility lever. Publishing one will not harm you, but it will not buy AI Overview citations by itself.
What still moves the needle
Indexable HTML with the answer in the first fetch. Clear titles, canonicals, and sitemaps. Fresh, specific pages that match real questions. Entity clarity (Organization / product JSON-LD that matches visible copy). Performance that does not punish humans.
Search Console’s generative AI performance reporting helps you see impressions in AI features — use it as measurement, not as a reason to invent markup Google ignores.
Where llms.txt still earns its keep
Coding agents, docs tooling, and some RAG-style workflows still benefit from a clean Markdown index of your product pages. AuditHQ’s AI-ready score checks reachability for agents and answer engines that fetch HTML or follow that convention.
Treat llms.txt as agent navigation and optional citation hygiene for non-Google surfaces — never as a substitute for SSR content and technical SEO.
A sane audit order for 2026
1. Fix crawl and index basics (robots allow important paths, sitemap, canonical host). 2. Make money-page copy visible without JavaScript. 3. Decide AI search bot policy deliberately. 4. Keep or add llms.txt for agents if you have product/docs surfaces. 5. Re-measure on a schedule — Google’s AI surfaces and your deploys both move.
AuditHQ website audits report classic SEO, performance, security on owner URLs, and AI-ready checks side by side so you do not confuse a Google lever with an agent file.