5.Fix what AI reads.

Here is the mental model that makes GEO tractable: an assistant describes you from what it can read. Change the reading, and you change the description. The reading list has two territories: pages other people control, and pages you control. Both are workable.

Territory one: the sources engines lean on

For any market, the engines keep returning to a surprisingly small set of independent sources: review platforms, directories, industry publications, best-of lists, comparison articles. These are the authority citations behind most answers, and they are discoverable: collect enough answers in your category and count which domains get cited.

That count is a work list. If the three sources your category’s answers lean on most do not mention you, being added to them is the highest-leverage move available: one listing can surface in thousands of future answers. If they mention you wrongly or thinly, correcting and enriching those listings is next.

This is also where wrong facts get fixed. Chapter four’s bad pricing and stale descriptions live somewhere specific: an outdated directory entry, an old article, a forgotten profile. Fix the source and the answers follow it, on the engines’ timetable rather than instantly, but they follow.

Territory two: your own site

Your site is the one source you fully control, and most sites are accidentally hostile to machine readers. The fixes are unglamorous and effective:

• Answer the actual questions. Buyers ask assistants full questions; pages that state the answer plainly, near a heading that matches the question, are the pages engines quote. Marketing prose that circles the point loses to a competitor paragraph that lands it.
• Put facts in text. Pricing locked in images, features implied by screenshots, and claims spread across animations are invisible to a machine reader. If a fact matters, it should exist as a plain sentence somewhere.
• Let the readers in. Many sites block AI crawlers wholesale, then wonder why assistants describe them badly or not at all. Decide deliberately which bots you admit; blocking retrieval crawlers removes you from live answers today.
• Explain the site to machines. An llms.txt file summarizes what your site is and where the important pages live, in the plain structured form engines parse best. It is one honest piece of the practice, not a magic switch.

The order of operations

Work the territories in evidence order, not effort order. Fix the wrong facts first, because misinformation compounds. Then close the gaps on the sources your category’s answers already lean on. Then make your own pages quotable, starting with the questions you are losing. Each of these is a discrete task with a before and after, which is what makes the next chapter’s scoreboard meaningful.