The buyer asks for the best tool for the job. An answer can only say what a product does if some page does.

Best tool for, X versus Y, alternatives to: what each question asks of your site

"Best tool for [a job] for [a kind of company]" asks for a category and a fit. The AI assistant needs a sentence that says what the product is and who it is for, in the buyer's words. "A platform for modern teams" is not a category, and a product without a category is not in a list of the best tools in one.

"X vs Y" asks for a comparison. An assistant can only build one from whatever pages compare the two: yours, the other product's, a reviewer's, a forum thread. A comparison page on your own site, honest about where the other product is stronger, is a page it can quote. Without one, it can only quote someone else's.

"Alternatives to X" asks for a category and a difference: cheaper, simpler, self-hosted, a free plan, a region where the data stays. A page that states the difference in one checkable sentence is the page; a page that says "the modern alternative" states no difference at all.

What none of the three can use: a gated demo, a pricing page that says "contact sales" on every tier, a features page made of verbs. A claim about how many teams trust you cannot be checked and is not what decides a match. The category sentence, the price and the integration list are.

What a software company's site has to state for an assistant to categorize it

Your pages reach the assistant as served HTML, fetched the way an AI crawler fetches them; OpenAI's documentation describes ChatGPT-User opening a page when a user asks about a product. What renders only after your framework's scripts run, it may never see.

SoftwareApplication is the schema.org type for the product, with applicationCategory, operatingSystem and an Offer carrying the price or the entry point. Organization is the type for the company, with the fields Google's organization documentation lists: name, url, logo, sameAs, contact points. Google recommends JSON-LD and says to mark up only what a visitor can see, so the pricing page and the Offer have to agree. One caution: Google's software app documentation requires an aggregateRating or a review on SoftwareApplication for its rich result, and reports the markup as an error without one. Add ratings only if customers gave them on your own site: Google's review snippet rules say not to aggregate ratings from other websites, so a score copied from a review platform does not qualify, and one you wrote never does.

Google's guidance on AI features answers the temptation to write separate pages for assistants: it lists no additional requirements for AI Overviews and AI Mode, no special markup or file, and one condition, that a page be indexed and eligible for a snippet. One set of pages, readable by a machine, is the whole instruction.

What our crawl found, and which findings carry over to a software site

The sample behind our audit of 9,483 small business websites was fetched by a crawler that ran no JavaScript and obeyed robots.txt, three pages a site, in 40 trades and 90 metros between late August and early September 2026. The sample was built around local trades and the audit publishes no separate cut for software companies, so read the figures as the state of small business sites, not of your category; the method section says how the sample was drawn.

No question-and-answer content at all on 61.5% of them; a question heading with nothing liftable under it on 50.7%. On a software site the shape is a features page of verbs and a pricing page with no numbers.

17.8% had no schema.org markup at all; of the 82% that did, most described the website, which on a software site means the framework's default WebSite and BreadcrumbList nodes and no SoftwareApplication or Organization. 13.7% had business schema missing the fields that would make it usable.

3.0% rendered their content only in JavaScript, so a crawler that runs no scripts saw an empty shell. A marketing site built as a single-page application is the software trade's version of this. Google documents rendering as a separate, deferred step, and nothing in the assistants' crawler documentation promises a rendering step at all.

7.2% sat behind bot protection that served the crawler a challenge page, and 3.2% disallowed AI crawlers in robots.txt on purpose. OpenAI documents that disallowing OAI-SearchBot removes a site from ChatGPT search answers and that disallowing GPTBot does not; a GPTBot line added in 2023 governs training data, not search answers. The free crawler access check reads a robots.txt token by token, which is the only way to see whether your answer-side rules say what you think.

The probe's questions were local ("best electricians in Brampton" and the like), so we do not apply its 14.9% named rate to software products, whose buyers ask in a different shape. What carries over is the mechanism: on inspection, the businesses that were named were the ones whose sites had answer-shaped content and a consistent identity.

What to fix first, in order

1. Write the category sentence (what the product is, for whom, in the buyer's words) and put it first on the homepage and in the meta description.
2. A pricing page in HTML with tiers, or the model and the entry point, and an Offer node that matches it.
3. One page per use case and one entry per integration, by name, saying what syncs which way.
4. One comparison page per named competitor, in checkable facts, dated, honest about where the other product is stronger.
5. SoftwareApplication and Organization in JSON-LD, with sameAs to the profiles that are yours, describing only what the page shows.
6. Public documentation in plain sentences, a dated changelog, and a security page that names the certifications you hold, as text.
7. Check robots.txt for the answer-side tokens (OAI-SearchBot, PerplexityBot and the rest, not only GPTBot) with the crawler access check, and check the served HTML with scripts off.

Being in the shortlist is the assistant's call, made where no vendor or agency can reach. The list above makes your product one it can categorize and quote; a fixed panel of buyer questions, asked monthly, is how you see whether it does.

Sources: our audit of 9,483 small-business sites, with its method · Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024) · Google, AI features and your website · Google, organization structured data · Google, introduction to structured data · Google, JavaScript SEO basics · OpenAI, crawler documentation · schema.org, Organization · schema.org, Offer