How AI search tools decide which software to suggest
Last updated
No AI company publishes how its assistant picks the software it suggests. What they do publish is the process, step by step. This guide sets it out, and what follows from it.
What the companies document
- It decides whether to search, then writes from what it found. Of Gemini, Google says “The model analyzes the prompt and determines if a Google Search can improve the answer”, runs “one or multiple search queries”, and then “processes the search results, synthesizes the information, and formulates a response” (Grounding with Google Search). OpenAI says of its own models that “the model can choose to search the web or not based on the content of the input prompt” (OpenAI’s web search tool).
- One question becomes several searches. Google’s AI Overviews and AI Mode “may use a ‘query fan-out’ technique”, which it describes as “issuing multiple related searches across subtopics and data sources” (AI features and your website). OpenAI’s reasoning models can “analyze results, and decide whether to keep searching”.
- For Google, the pool is its search index. Google describes its AI features as “relying on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index” (Google’s guide to generative AI search). To be a supporting link, a page “must be indexed and eligible to be shown in Google Search with a snippet”.
- For ChatGPT search and Perplexity, their crawlers have to be let in. “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers” (OpenAI’s crawlers). Perplexity says PerplexityBot “is designed to surface and link websites in search results on Perplexity” (Perplexity’s crawlers).
Each of these describes one company’s own product. None says how one source is weighed against another, and nobody outside those companies knows.
What follows for a software company
- Be in the index behind each assistant. Google’s, for Gemini and AI Overviews. For ChatGPT search and Perplexity, let OAI-SearchBot and PerplexityBot in. How to check.
- Have a page for the narrower question. If one question becomes several searches, the page that gets found is the one that answers a part of it: your pricing, one use, one comparison.
- Find the pages your market’s answers read, and be on them. They differ by market. Ask your buyers’ questions and read the sources.
- Expect a short list. The question is who is on it. Why they are usually the established ones.
- Measure it, because it moves. The same question can get a different answer tomorrow. How to measure.
Wellread shows which sites the answers in your market drew on, and who they named. What an audit is.