Guide · Windows apps

How do Windows apps get recommended by ChatGPT and AI search?

Updated 2026-11-11 · kagentx

Short answer

Write your website's Windows download page in full: a one-sentence definition, system requirements, price, version and install command, and make the Microsoft Store description and the winget ShortDescription say the same thing. Tested on 2026-11-11: Microsoft Store's robots.txt has no restrictions at all, and an app page's title, description and SoftwareApplication data can be read without running scripts, but the page body needs script rendering; Chocolatey allows search crawlers and blocks training crawlers; Softpedia blocks PerplexityBot among others. Whether AI cites these pages we have no evidence (not verified).

Do the channels for Windows apps let AI crawlers in?

On 2026-11-11 we read the robots.txt of the common channels for Windows apps one by one and judged them for seven User-Agents: GPTBot (OpenAI training), OAI-SearchBot and ChatGPT-User (ChatGPT search and user-initiated visits), ClaudeBot, PerplexityBot, Google-Extended (Google's AI training switch) and Bingbot. The rule is that a named group wins, and without one the * group applies. This only shows how robots.txt is written. It does not show that those crawlers actually came, or that AI cited these pages (we have no evidence of citations: not verified).

ChannelGPTBotOAI-SearchBot / ChatGPT-UserClaudeBotPerplexityBotGoogle-ExtendedBingbot
Microsoft Store (apps.microsoft.com; robots.txt has only a Sitemap line)AllowedAllowedAllowedAllowedAllowedAllowed
GitHub (where winget-pkgs and your release pages live)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)
Chocolatey (community.chocolatey.org)BlockedAllowedBlockedAllowedBlockedAllowed
Softpedia (www.softpedia.com)BlockedAllowed (only some paths restricted)BlockedBlockedAllowed (only some paths restricted)Allowed
MajorGeeks (majorgeeks.com; only one download address is blocked)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)Allowed (only some paths restricted)

How to read this table: for the same app, AI search crawlers are treated differently on different channels. Chocolatey allows search crawlers and blocks training crawlers; Softpedia blocks even PerplexityBot, and its app page returned 403 to our curl, so the page content is not verified; Microsoft Store has no restrictions at all. For how to write robots.txt and how the crawlers differ see How to write robots.txt so AI crawlers can get in.

Conclusion: not every Windows directory can be read by AI search, and the only place you fully control is your own website. Your download page should state system requirements, price, version and install command in full, and the pages on other channels should stay consistent with it.

What can crawlers read on a Microsoft Store app page?

We fetched the Windows Terminal store page (2026-11-11, HTTP 200, about 228 KB). Without running scripts you can read the title, the description and a SoftwareApplication structured data block; the page body is rendered by script, leaving only 106 characters once scripts are removed, and the <noscript> says JavaScript is required.

FieldCan it be read without running scripts?What we measured
Title <title>YesApp name plus store template text plus Microsoft Store
meta description / og:descriptionYesThe full app description, matching what you entered in the store
SoftwareApplication JSON-LDYesname, description, image, url, author, applicationCategory, operatingSystem (Windows 10), offers (price 0), 3 screenshots
Rating (aggregateRating)YesRating 4.6 from 44 reviews, in the JSON-LD
canonical and hreflangYescanonical carries hl and gl parameters; hreflang covers 47 languages
Visible page body (description, features, release notes)NoThe body is rendered by script; the <noscript> only asks for JavaScript to be enabled

So the most reliable parts of a store page are the title, the description and the JSON-LD: write the first paragraph of the store description well and it appears in all three. Release notes and feature lists live only in the script-rendered body, which crawlers may not read, so write them again on your website. For whether AI crawlers run scripts see Do AI crawlers render JavaScript?. We did not read the official store field rules (Partner Center): not verified.

Which fields in a winget manifest can people and crawlers see?

A winget manifest is a public YAML file in the microsoft/winget-pkgs repository on GitHub, readable by anyone (on 2026-11-11 we read Windows Terminal's manifest, which opens both on raw.githubusercontent.com and on the github.com file page). Microsoft provides no public product page for it, and whether third-party sites generate pages from it or AI cites it is not verified. The official docs (Microsoft Learn, viewed 2026-11-11) and what we saw line up below.

manifest fieldDocsHow to write it
PackageName, PublisherRequired: the app name and the publisherExactly the same as the name on your website and Microsoft Store
ShortDescriptionRequired: a description of the appOne sentence: what it does, for whom; same meaning as the first paragraph of your website; no keyword stuffing
TagsIn the example; Windows Terminal's manifest has 12 tagsWrite the words users search for, no more than about ten
MonikerOptional; users who search this word can find the appA short, memorable name that does not clash with another app
PackageUrl, ReleaseNotesUrlIn the examplePoint to your website and changelog page, not the home page
LicenseRequiredConsistent with the license in your repository
PackageIdentifier: ExampleCo.ExampleWin
PackageVersion: 2.3.0
PackageLocale: en-US
Publisher: ExampleCo
PackageName: ExampleWin
PackageUrl: https://example.com/
License: MIT
ShortDescription: One sentence: what it does and who it is for.
Tags:
- screenshot
- productivity
ReleaseNotesUrl: https://example.com/changelog/
ManifestType: defaultLocale
ManifestVersion: 1.10.0

The above is an example of how to write it (URLs and names are placeholders). The manifest, the store page and your website must agree with one another: name, description, version and URL; where they do not match, AI may mix them up when answering.

What can be read on a Chocolatey package page?

We fetched the 7-Zip package page (2026-11-11, HTTP 200, about 235 KB): the body is server-rendered, and the download counts, last update, maintainers, tags, full description, release notes link and dependencies are all in the HTML; there is no structured data. The og:description is the package's one-sentence description and the meta description is Chocolatey's generic introduction, so the snippet in search results most likely comes from the page body.

How do you make a Windows app easier to recommend, step by step?

  1. 1. First make sure your website does not block AI search crawlers

    Use AI crawler check to look at robots.txt. Do not block OAI-SearchBot, PerplexityBot or Bingbot; GPTBot and Google-Extended are training crawlers, so blocking them is your choice.

  2. 2. Make a Windows download page with everything on it

    First paragraph: a one-sentence definition; then system requirements (Windows 10 or 11, x64 or ARM64), price, current version, update date and install methods (installer, Microsoft Store, the winget install command). For how to write the price see How should a pricing page be written so AI answers your price correctly?.

  3. 3. Add SoftwareApplication structured data

    Write the minimum system in operatingSystem and the current version in softwareVersion; for how to write it see Should a SaaS add SoftwareApplication structured data?. Example (URLs are placeholders):

    {
      "@context": "https://schema.org",
      "@type": "SoftwareApplication",
      "name": "ExampleWin",
      "applicationCategory": "UtilitiesApplication",
      "operatingSystem": "Windows 10, Windows 11",
      "softwareVersion": "2.3.0",
      "downloadUrl": "https://example.com/download/",
      "releaseNotes": "https://example.com/changelog/"
    }
  4. 4. Make the first paragraph of the store description, the manifest's ShortDescription and the first paragraph of your website say the same thing

    The store page's title, description and JSON-LD all come from the start of the store description; when you change one, check the other two.

  5. 5. Keep the changelog on your website

    The store page's release notes are rendered by script and crawlers may not read them; for your website's changelog page see Do changelog pages help AI search?.

  6. 6. Test again after a few weeks with the same questions

    Ask ChatGPT and Perplexity “what apps on Windows do X”, three times per question, and note whether you are mentioned and which URLs are cited; for the Mac channels compared see How do Mac apps get recommended by AI?.

FAQ

Will ChatGPT cite a Microsoft Store page?

We found no official statement and no evidence of actual citations (not verified). What we can confirm: on 2026-11-11 apps.microsoft.com's robots.txt has no restrictions, and an app page's title, description and JSON-LD can be read without running scripts.

Should I publish on both Microsoft Store and a direct download from my website?

The choice of channel depends on your product and users, not on AI. Whichever you choose, have one page on your website that states system requirements, price and install methods in full, and links to the other channels.

Is a winget manifest useful for AI search?

The manifest is a public file; keeping its fields (name, description, tags, URLs) consistent with your website helps users and third-party sites. Whether AI cites it directly we have no evidence (not verified).

Chocolatey blocks GPTBot; does that affect ChatGPT search?

According to OpenAI, GPTBot is a training crawler and ChatGPT search uses OAI-SearchBot. On 2026-11-11 Chocolatey's robots.txt does not name OAI-SearchBot, so the * group applies and it is allowed.

Are download sites like Softpedia and MajorGeeks worth it?

We only verified their robots.txt: Softpedia blocks GPTBot, ClaudeBot and PerplexityBot, and its app page returned 403 to curl, so its content is not verified; MajorGeeks blocks only one download address. Whether they are worth it depends on whether your users look for software there, not on AI.

Will AI read the rating on a store page?

On the Windows Terminal store page we fetched on 2026-11-11, the rating (4.6 from 44 reviews) is in the JSON-LD and can be read without running scripts; whether AI uses it is not verified.

Does a Windows app need to state ARM64 support?

We suggest it. The JSON-LD on the store page we fetched gives only Windows 10 for operatingSystem and no chip information; when users and AI answer “does it run on ARM PCs”, they need one clear statement, and your download page is the place you control.

Sources

kagentx

Can crawlers read your pages?
Let the evaluation tell you

The evaluation checks whether crawlers can get in and whether the text is served directly, then asks ChatGPT, Perplexity and Claude (with web search on) your buyers' questions and sees whether you are mentioned. The Basic plan is $1, and results usually take 2–3 minutes.