--- title: "The model names the doc it can lift." description: "Search for developer tools and AI companies: docs, integration pages, and being named in LLM answers. SEO, AEO, GEO, and AIO." canonical: https://shoutrank.com/industries/developer-tools markdown: https://shoutrank.com/industries/developer-tools.md --- # The model names the doc it can lift. Search for developer tools and AI companies: docs, integration pages, and being named in LLM answers. SEO, AEO, GEO, and AIO. A developer asks how to connect your SDK to the stack they already run. If the only working example lives in a collapsed app or a PDF, the answer names the library with a copy-paste README. ## Marketing ranks for the brand. The answer quotes the docs. AI companies and developer tools get named when a model can lift a task: install, authenticate, call the endpoint, handle the error. A marketing site that describes “the platform for builders” gives the model nothing to paste. The competitor’s docs do, so the competitor is the name in the reply. Integration pages are the other miss. Stripe, Next.js, Postgres, a vector store: each pairing is a query. One “integrations” grid with logos is not a page about any of them. Changelogs and versioned docs then compete with the stable task page, and the model cites a stale snippet. ## A task, a stack, and a partner. The unit is a doc or an integration, not a campaign landing page. ### Task stream a chat completion in the node SDK The title is the task. The first screen is a working sample, the auth step, and the error a new user hits. ### Stack use the SDK inside a typescript monorepo Package name, version, and the file it belongs in. A model cannot cite a diagram. ### Integration send traces to an existing OpenTelemetry collector One partner, one page. The logo wall does not answer the pairing. ### Named in a list vector databases people use for retrieval You get named if the docs state what you store, what you do not, and how a call looks. A manifesto will not be the citation. ## HTML docs a model can fetch. The work is the stable task page and the integration, kept in step with the current version. ### Task pages in HTML Each frequent job has a URL with the sample in the document, not only inside a client-rendered playground. The playground can stay; it cannot be the only copy. ### One page per integration Partner, version, and the three steps. When the partner’s API changes, this page changes or it is labeled for the old version. ### A current version, clearly The latest stable doc is the one we want cited. Older versions stay reachable and marked, so a model does not teach last year’s method as current. ### Names that match the package The npm or PyPI name, the product name, and the company name are explicit on the doc. Models mix them up when the page never lines them up. ## Cited domains github.com is the most cited domain in Developer tools in this snapshot. Source for this chart: [Top cited domains](https://shoutrank.com/data-studies/top-cited-domains.md). This data is accumulated from multiple sources and internal training data, which helps us get the client cited. | Domain | AI Overview | ChatGPT | Gemini | AI Mode | Perplexity | | --- | ---: | ---: | ---: | ---: | ---: | | github.com | — | 88 | 0 | — | 1 | | reddit.com | — | 119 | 0 | — | 0 | | atlassian.com | — | 0 | 1 | — | 1 | | en.wikipedia.org | — | 65 | 0 | — | 0 | | g2.com | — | 0 | 0 | — | 1 | | gitnux.org | — | 0 | 0 | — | 1 | | itechguides.com | — | 0 | 0 | — | 1 | | joinhiper.com | — | 0 | 0 | — | 1 | | scrunch.com | — | 0 | 0 | — | 1 | | spacelift.io | — | 0 | 0 | — | 1 | | topickz.com | — | 0 | 0 | — | 1 | | docs.aws.amazon.com | — | 47 | 0 | — | 0 | A dash means this snapshot has no count for that engine. A zero means the domain was not cited in the measured sample. ## What is at stake - **What gets named.** The doc or integration a model can paste, not the campaign site. - **URL that should own it.** A task page in HTML, with the current version on it. - **What fails quietly.** A playground, a PDF, or a changelog taught as the method. ## The answer pastes someone else’s README. A developer asks how to stream a completion or ship traces to the collector they already run. The marketing site says “built for builders.” The rival’s docs show the call, the auth step, and the error. The model names the rival. ## Stream a chat completion from the Node SDK The title is the task. The first screen is code a developer can run, in HTML a crawler can read. - Install the current package. The sample on this page targets that version, not last year’s method. - Create a key, pass it as an environment variable, and stream tokens to stdout. - If the first call returns 401, the key is in the wrong header. The page shows that error beside the sample. ## Sequence 1. **Pick the tasks.** The jobs people already ask: install, authenticate, one hard integration. Not every endpoint. 2. **Serve them as HTML.** The sample lives in the document. The playground can stay. It cannot be the only copy. 3. **Split the partners.** Each real integration gets a page with a version. The logo wall stays a directory. 4. **Mark the old method.** Previous versions stay reachable and labeled, so a model does not teach them as current. We will not add an llms.txt file and call the product citable while the task still lives only inside a client-rendered shell. ## What we watch Fetches of the task and integration URLs, whether model answers paste your sample or a competitor’s, and how often a cited snippet is from an old version. A named result is published when that client clears the figure. A client figure is added here only after that client clears it. ## Developer tool questions ### Why do the docs beat the marketing site in model answers? Because the question is a task. A model quotes the page that shows the call. We put the sample on a fetchable HTML URL and link it from the marketing page, instead of hoping the homepage is the citation. ### Does every integration need its own page? Every pairing people already search or ask about. A logo does not. If you have forty partners and four real queries, we write the four and keep the rest as a directory that points at them. ### What if the docs are a single-page app? Then many crawlers and models see an empty shell. We serve the task content as HTML, or as a static export, and we check the audit to confirm the sample is in the response. ### Will an llms.txt file get the product named? It can point a model at the URLs you want read. It does not replace a page that answers the task. We add it after the task pages exist, not instead of them. ### How do changelogs stay out of the way? Release notes stay release notes. The task page is updated in place and shows the version the sample targets. We do not let a changelog outrank the method. ## Audit the doc you want pasted. Run one task URL or one integration URL through the free visibility audit. The score says whether that sample can be fetched at all. Book a visibility audit: [Contact](https://shoutrank.com/contact.md).