AI search research and advisory
We study how AI assistants find, read, and cite websites.
Modo AI publishes plain-language research on ChatGPT, Google's AI Overviews and AI Mode, Perplexity, Claude, and Copilot, and offers advisory work for teams that want to see how those systems treat their own site.
Advisory work
Engagements are scoped to the questions a team needs answered about its own site, and they rest on the same primary sources and methods as the published guides.
Visibility audits
A one-time check of whether AI crawlers, fetchers, and answer engines can reach and read a site. It covers robots.txt groups, CDN bot settings, rendering, and indexing, and it samples answers for a defined set of questions.
Strategy
Working out which questions a site should answer, which pages carry each one, and what those pages need to state plainly, with each recommendation tied to what the AI companies have published about their systems.
Monitoring
Repeated runs of a fixed question set across engines, read next to Search Console and Bing Webmaster Tools data, and reported as ranges because answers vary from one run to the next.
Where to begin with AI search
Most AI answers that carry links start as a search. Google says AI Overviews and AI Mode are built on its core ranking systems and can only link to pages that are indexed and eligible for a snippet, and OpenAI says ChatGPT rewrites a question into searches of its own and sometimes sends them to outside search providers. Our guide to how AI search engines choose and cite sources follows a page through that process, and the essay Ranking for machines makes the longer case for writing with machine readers in mind.
Access is the next thing to check. OpenAI, Anthropic, and Perplexity each run one bot for their search index and another for fetches a user triggers, and a robots.txt group, a CDN bot setting, or a page that only renders with JavaScript can keep a site out of reach. The crawler and agent access guide covers each of those checks.
Measurement is the hard part, because the same question can return different sources from one run to the next. The measurement guide explains what Search Console, Bing Webmaster Tools, analytics, server logs, and prompt tracking can and can't show, and the AI search glossary defines the terms that come up along the way.
Guides
All guidesHow AI search engines choose and cite sources
Query fan-out, where the candidate pages come from, and why engines cite fewer pages than they read.
Making a site readable to AI crawlers and agents
Which bot does what, how robots.txt groups apply, Google's separate controls, CDN settings, JavaScript, and llms.txt.
03Measuring AI search visibility
What Search Console, Bing Webmaster Tools, analytics, logs, and prompt tracking can each show, and where each one goes blind.
04AI search glossary
Short definitions for the terms that come up in AI search work, with links to the guides that cover them in depth.
Recent research
All researchRanking for machines
What AI search agents read when they visit a site on a user's behalf, and how structure, specificity, and source authority shape what they take away.
The protocol layer between agents and tools
A shared protocol for agent and tool communication is emerging, and it changes how services get discovered.
Authentication when the user is a machine
OAuth was designed for people clicking Allow. AI agents need to authenticate without a person in the loop.