Ask ChatGPT or Perplexity for the best tool in your category and you get a short list with reasons attached. That list is assembled from pages the model can read and trust. For most software companies the failure is upstream of any content strategy, in the rendering: the crawler requests the page, gets a shell with a script tag, and moves on. Everything downstream is wasted if that is where you stop.
1. Return the content on the first response
Googlebot will generally execute your JavaScript, eventually. The AI crawlers largely will not; most take the first HTML response and go. A client-rendered marketing site is therefore roughly visible to Google and roughly invisible to the models. Server-rendering the marketing surface is the single change with the largest effect, and it is usually the cheapest thing on this list because it is a build-configuration problem rather than a content problem. This was the entire North Stack rebuild: a six-page bundle became 34 edge-rendered URLs.
2. Say what the product is in words on the page
Models synthesise positioning from prose, not from a hero animation. Somewhere in readable text you need the category you are in, who it is for, what it replaces, what it costs, and where the boundaries are. Teams that write like this rank in AI answers for unglamorous reasons: the model has a sentence it can quote. Vague aspirational copy gives it nothing, which is why so many well-designed sites lose to a competitor with a plainer one.
3. Build the comparison and integration surface
A large share of category demand is phrased as a comparison: alternatives to X, X versus Y, does it work with Z. If you have a page for each, you are in the answer for each. If you do not, the answer comes from a review aggregator or a competitor's comparison page, and both of those are written by someone with an interest in the outcome. Generating these from one data file means covering the whole surface in a build rather than one page a month.
4. Give agents a clean version of the page
Beyond HTML, it helps to serve a plain-text or markdown representation of each page at a predictable URL, plus an index of your key pages at the site root. Agents that support it will take that path and get exactly the content you intended, with no layout to strip. We built this for North Stack and it is now the default on our software builds.
5. Structure the data the answer needs
Pricing, plan tiers, supported platforms, integrations and FAQs should be present as structured data as well as prose, so a model reading your pricing page does not have to infer it from a table it half-understands. This is where AI answers get specific enough about your product to be persuasive rather than vague.
Measuring it
You watch crawler behaviour rather than guessing. Cite-met, which is our own product, tracks which AI crawlers hit which pages, how often, and how the trend moves. In practice the crawl volume moves within weeks of a rendering fix, and the citations follow over the next quarter. Our own site now takes 32% of its traffic from AI engines, over 11,618 visits in a single week.
Where to start
The free audit fetches your pages the way a crawler does and shows you what comes back, which structured data is present, and which comparison searches you are closest to winning. It lands in 48 hours whether or not you hire us.