AI assistants do not publish exactly how they pick sources, and methods differ by product and change over time. Many can retrieve live web pages for some questions, while others rely on what they learned in training. It is reasonable to expect that clear, relevant, trustworthy, accessible pages are easier to use and cite, but no one outside can prove a formula.
- Source selection is not public, differs between assistants and changes often.
- Some assistants search the web live; others lean on learned knowledge.
- Access, clarity and credibility are sensible foundations to build on.
- Be sceptical of anyone claiming to know exact ranking rules.
Do AI assistants search the web or just remember?
Both patterns exist. A language model learns patterns from large amounts of text during training, which gives it general knowledge up to a point in time. Many assistant products also add a search or retrieval step that fetches current web pages and uses them to write an answer, sometimes showing links.
Which pattern applies depends on the product, the settings and often the question itself. A request for a recent fact or a local supplier is more likely to trigger a lookup than a request to explain a basic concept. This is why a single universal ranking recipe does not exist.
What do we actually know?
Little is documented in detail. The companies describe their products in general terms, and their documentation about crawlers and web access is the most concrete public information. Beyond that, most commentary is inference from observation, and observations can become outdated within months.
A useful habit is to separate what is stated by the providers, what is observed by testing and what is merely assumed. Stated facts, such as which crawlers exist and how to allow or block them, are reliable. Claims about hidden weighting deserve caution, especially when they come with a sales pitch.
- Stated: published crawler names and access controls.
- Observed: patterns from testing, which may change.
- Assumed: hidden weightings no one outside can confirm.
- Always check current official documentation.
What makes a page easier to use and cite?
Reasonable assumptions follow from the task. A system that must write a correct answer in a short time prefers pages it can fetch, parse and interpret quickly: plain readable text, clear headings, direct answers, and specific details such as who, where and under what conditions. Pages that bury the answer, hide content behind scripts or contradict other sources are harder to rely on.
Credibility cues likely help as well: identifiable authorship, a real organisation behind the site, consistent facts across other sites and content that is current and corrected when wrong. These are the same cues a careful person uses to decide whether to trust a source.
- Content available in readable text, not only images.
- Clear headings and direct, self-contained answers.
- Specific, accurate and dated information.
- Identifiable authors and a real organisation.
How much does the rest of the web matter?
A great deal, probably. Assistants form impressions from many sources, so what directories, review sites, news mentions and partner pages say about you can shape how you are described. If your own site says one thing and widely used listings say another, the answer may blend or hedge.
That is why consistent business facts across the web, genuine reviews and credible mentions deserve the same attention as your own pages. It is less glamorous than clever tactics, and considerably more durable.
Can I control whether AI crawlers see my site?
Yes, to a degree. Providers publish the names of their crawlers, and rules in your robots.txt file can allow or disallow them. Your hosting or security settings may also block automated traffic without you realising, which can accidentally hide your site from useful tools.
Make that decision deliberately. Some publishers block AI crawlers to protect content; many businesses that want to be found prefer to allow them for public pages. Check each provider's current documentation, since names and behaviours change.
What should I do with this uncertainty?
Build for the durable principles and test rather than theorise. Keep pages clear and accessible, align your facts, earn real reputation, and record what assistants say about you using a fixed set of questions. Adjust as the evidence changes.
Treat any confident claim about how a specific assistant ranks sources as a hypothesis. The honest position in this young field is that nobody can guarantee inclusion, and the best preparation is being a source worth citing.
Frequently asked questions
Do all AI assistants use the same sources?
No. Products differ in whether they search live, what they index and how they present sources, and these behaviours change over time.
Does ranking first on Google mean an assistant will cite me?
Not necessarily. Strong search visibility may help in assistants that retrieve from search results, but there is no guarantee for every product.
Can I see which pages an assistant used?
Some products show citations or links for certain answers. Others do not, so you may need to test and infer.
Is paid placement possible in assistant answers?
Advertising formats are evolving in some products, but organic mentions cannot be bought. Read any provider's current policies for specifics.
How often should I test what assistants say?
Every month or two with a fixed question list is sensible, plus after major site or reputation changes.
Need help with this? See our GEO — Generative Engine Optimisation service or talk to Yash Parikh.