Building for generative engines, a practical checklist

The two to seven sources an AI actually cites

When a language model answers a question, it typically draws on a small, specific handful of sources rather than the ten results a traditional search page would show. Being one of them requires a level of structural clarity that most websites, built for people skimming quickly rather than machines parsing precisely, simply do not have by default.

What we did on our own site

Before recommending any of this to a client, we applied it to ourselves. We added structured schema for our organisation, our services and every case study we publish, wrote a plain llms.txt file that states directly who DigiIQ is and what we do, and went through every page checking that a fact stated in one place agreed with the same fact stated anywhere else on the site.

The llms.txt file in particular is a small, almost old fashioned idea, a plain text file that simply tells an AI system what a site is and how to understand it, but it is one of the few places where a brand gets to describe itself directly to a model rather than hoping the model infers it correctly from marketing copy.

The checklist

State plainly what a page is about in the first two sentences, before any scene setting or brand voice. Use headings that describe what a section actually says rather than teasing the reader into scrolling further. Add schema for organisation, articles and services wherever they apply. Publish an llms.txt file, and keep it accurate as the business changes. Most importantly, keep facts, numbers and claims consistent across every page that mentions them, since a language model that finds two different answers to the same question usually hedges or skips the source entirely rather than guessing which one is correct.

The mistake this checklist will not fix

None of this works if the underlying content has nothing worth citing in the first place. Structure earns a model’s attention, it does not manufacture expertise that was not there to begin with. The brands that benefit most from this practice are the ones who already have something specific and true to say, and simply were not saying it clearly enough to be heard.