Find AI Companies by Niche or Problem on artificialintelligencecompanies.com

Purpose

Given a niche (e.g. "healthcare", "customer service", "legal") or a problem statement (e.g. "automate inbound phone calls", "label training data", "detect insurance fraud"), return matching AI companies and startups from the public directory at artificialintelligencecompanies.com. For each match, return the company name, canonical directory URL path, and a short description. Read-only; never mutates the directory.

The site is explicitly agent-friendly: robots.txt allows GPTBot, ClaudeBot, ChatGPT-User, PerplexityBot; /llms.txt advertises the API; and /.well-known/agent-manifest.json enumerates capabilities. No API key is needed for read access.

Prompt assumption (placeholder template): the source prompt referenced {niche} and {problem} as unfilled placeholders. This skill treats them as two synonymous facets of the same input — the caller supplies one free-text string that is either an industry/vertical niche or a problem-to-solve phrase. The skill routes the same query through the same endpoint either way; the directory's /api/search/ does keyword matching across both category names and company descriptions.

When to Use

Mapping a vertical (healthcare AI, legal AI, computer vision, AI receptionists, training-data services) to a shortlist of vendors.
Translating an unstructured pain point ("our reps are drowning in tickets" → AI customer service vendors; "we need someone to phone-screen leads" → AI receptionist vendors) into a candidate vendor list.
Building a market-map / competitive-landscape dataset across one of the 12 directory categories.
Anywhere you'd otherwise scrape an AI vendor list by hand — the JSON API is faster and cheaper than rendering the site, and /cat/<slug>/ pages have richer JSON-LD than the API returns.

Workflow

The site exposes a documented read-only REST API plus an MCP server, both linked from /llms.txt. Lead with the JSON API for keyword queries, and supplement with the /cat/<slug>/ HTML pages when you need full (non-truncated) descriptions or a complete category roster. A browser is only needed for the HTML category pages — and even those can be fetched as plain HTML via browse cloud fetch <url> --proxies; no JavaScript renders the company list (it's server-rendered with JSON-LD embedded).

Pick the input shape. Keyword search (one query string covering the niche or problem) is the common case. If the caller already knows the directory's taxonomy slug (e.g. healthcare-ai), prefer the category page directly — it returns more data.
Keyword search (primary path) — when input is a niche/problem string:
```
GET https://artificialintelligencecompanies.com/api/search/?q={query}
```
- q is required, URL-encode multi-word queries (q=customer%20service).
- Returns {"companies": [{name, url, description}, ...], "categories": [{name, url, description}, ...]}.
- Hard-capped at 5 companies + matching categories regardless of how popular the term is. A limit query param is silently accepted but ignored — see Gotchas.
- Descriptions are truncated to ~100 chars with a ... suffix. Use step 4 to fetch the full text.
- Returns HTTP 200 with empty arrays ({"companies": [], "categories": []}) for no-hit queries; also returns 200 with empty arrays for 1-character q despite the OpenAPI declaring minLength: 2 — treat empty-arrays as no-match.
- Any matching categories[*].url (path like /cat/healthcare-ai/) is your hook into step 3 for the full vertical roster.
Category-roster search (best for full coverage of a vertical) — when input is a vertical/niche and you want the entire roster of vendors in that category:
```
GET https://artificialintelligencecompanies.com/cat/{slug}/
```
- Parse <script type="application/ld+json"> blocks; there are 3 per page. The third is @type: ItemList with each item as an Organization containing name, url, and a full untruncated description. The first is a CollectionPage with numberOfItems (use as a count of expected rows).
- Use this whenever the search API truncates a description you need, or when 5 results aren't enough.
- Stable 12-category enum (as of 2026-05-19, verified via /api/categories/?format=json): ai-automation-agencies, ai-consulting, ai-customer-service, ai-image-generators, ai-platforms, ai-receptionist, computer-vision, foundation-model-providers, healthcare-ai, legal-ai, machine-learning-platforms, training-data.
- To map a free-text vertical to a slug without keyword-searching first, hit GET /api/categories/?format=json (returns {success, total, categories: [{id, slug, name, description}, ...]}).
Hydrate full descriptions — when a search-API hit gave you a truncated description and the company belongs to a category, the company will also appear in that category page's JSON-LD ItemList with the full text. Cross-reference by name. (Direct individual-company JSON is not available — see Gotchas: /api/companies/{id}/ is 500, and /co/<slug>/ HTML is 500.)
Combine and dedupe. If your input is both a niche and a problem (e.g. "AI receptionist for plumbing companies"), run both phrasings through /api/search/?q= and union by name. Then for each company whose category you can identify (via the categories[] array in the same response), pull the full description from the corresponding /cat/<slug>/ JSON-LD.

Alternative: MCP server

/llms.txt advertises http://artificialintelligencecompanies.com/mcp-server/ as a downloadable MCP server. If you have an MCP-aware host (Claude Desktop, Cursor, etc.) and are doing repeated discovery queries, install it. Not exercised in this skill's iteration — recommended_method stays api because the raw HTTPS calls are zero-setup and equally effective for one-shot lookups.

Browser fallback

Only useful when the JSON API is fully blocked (no evidence this has ever happened — no auth, no anti-bot, served via Cloudflare with permissive robots.txt):

browse open https://artificialintelligencecompanies.com/search/?q={query} — the HTML search results page. Server-rendered, no JS required. Note: this page does not carry JSON-LD; you'd be parsing rendered HTML, which is strictly worse than the API.
browse open https://artificialintelligencecompanies.com/cat/{slug}/ — the category page. Server-rendered with rich JSON-LD as described in step 3.

browse cloud fetch <url> --proxies is sufficient for both pages — no headless browser is required. Stealth/verified flags are not needed for any path on this site.

Site-Specific Gotchas

/api/companies/ is broken site-wide: the OpenAPI-documented GET /api/companies/?search=&category=&country=&status= returns HTTP 500 for every parameter combination tested on 2026-05-19 (no params, ?search=healthcare, ?category=healthcare-ai, ?limit=3, with and without format=json). The browsable HTML form at the same path also renders the Django "Add Category Api" form rather than a list. Do not waste turns on this endpoint — use /api/search/?q= for keyword search and /cat/<slug>/ JSON-LD for category rosters.
/api/companies/{id}/ (detail by id) is broken: returns HTTP 500 for id=1 (and presumably all ids). The OpenAPI spec advertises it; the implementation is unavailable as of 2026-05-19.
Every /co/<slug>/ company detail HTML page returns 500: verified across openai, anthropic, abridge, databricks, zendesk, goodcall — 5 different company slugs spanning the alphabet. The site lists 208+ companies in its hero, but none of the canonical company-profile URLs render. The only place to read a full company description today is the /cat/<slug>/ page's JSON-LD ItemList (step 3) or the truncated search response.
/api/search/ is hard-capped at 5 company results: ?limit=N is silently accepted but ignored — no matter what value you pass, you get at most 5 companies[] entries plus matching categories. Treat the search endpoint as a "top hits" lookup, not a paginated list. For broad coverage, walk the relevant category pages instead.
/api/search/ descriptions are truncated to ~100 chars with ... suffix: full untruncated descriptions only live in the JSON-LD on /cat/<slug>/ pages. Cross-reference by name.
/api/categories/ requires ?format=json: without it the endpoint serves the Django REST framework browsable HTML form ("Add Category Api"), not JSON. The OpenAPI spec doesn't mention this. Always append &format=json (or send Accept: application/json if you control headers). The /api/search/ endpoint does not need format=json — it defaults to JSON.
minLength: 2 on the search q param is not enforced: ?q=x returns HTTP 200 with empty arrays rather than HTTP 400. Don't rely on the spec — validate input on your side if length matters.
url fields in API responses are relative paths, not absolute URLs (e.g. "url": "/co/anthropic/", "url": "/cat/healthcare-ai/"). Prefix with https://artificialintelligencecompanies.com before display. Also note the canonical URLs in /llms.txt, sitemap.xml, and JSON-LD use http:// (not https://) — the site redirects http:// → https:// cleanly, but normalize on https:// for cross-system consistency.
/cat/<slug>/ pages have 3 JSON-LD blocks: CollectionPage (metadata + numberOfItems), BreadcrumbList (navigation), and ItemList (the actual roster). Parse the third one — match on "@type": "ItemList", not on position, since order is not contractual.
Stable 12-category taxonomy as of 2026-05-19 (/api/categories/?format=json total: 12). New categories may be added — re-enumerate before assuming the list is current.
No auth, no anti-bot, no stealth required: Cloudflare-fronted, served from edge cache. --proxies and --verified flags on Browserbase sessions are unnecessary. The site explicitly opts in to AI crawler traffic (see robots.txt).
MCP server URL advertised but not exercised here: /mcp-server/ is listed in /llms.txt and /.well-known/agent-manifest.json but was not validated as part of this skill's iteration. If you are an MCP-aware client, fetching /mcp-server/ should give you tool definitions equivalent to the three working HTTP endpoints.

Expected Output

Shape A — keyword search returned matches

{
  "query": "customer service",
  "method": "api",
  "endpoint": "/api/search/?q=customer%20service",
  "matched_categories": [
    {
      "name": "AI Customer Service",
      "slug": "ai-customer-service",
      "url": "https://artificialintelligencecompanies.com/cat/ai-customer-service/"
    }
  ],
  "companies": [
    {
      "name": "IBM Watson",
      "url": "https://artificialintelligencecompanies.com/co/ibm-watson/",
      "description": "IBM Watson is IBM's enterprise AI platform providing machine learning, natural language processing, ...",
      "description_truncated": true
    },
    {
      "name": "Zendesk",
      "url": "https://artificialintelligencecompanies.com/co/zendesk/",
      "description": "Zendesk is a customer service and employee service software provider headquartered in San Francisco,...",
      "description_truncated": true
    },
    {
      "name": "Intercom",
      "url": "https://artificialintelligencecompanies.com/co/intercom/",
      "description": "Intercom is a customer service software provider headquartered in San Francisco with roots in Dublin...",
      "description_truncated": true
    },
    {
      "name": "Uniphore",
      "url": "https://artificialintelligencecompanies.com/co/uniphore/",
      "description": "Uniphore is a leading enterprise conversational AI company specializing in voice and vision AI solutions...",
      "description_truncated": true
    },
    {
      "name": "Goodcall",
      "url": "https://artificialintelligencecompanies.com/co/goodcall/",
      "description": "Goodcall provides agentic voice AI for inbound phone operations, including lead capture, appointment...",
      "description_truncated": true
    }
  ],
  "result_count": 5,
  "result_capped_at_5": true
}

Shape B — category-roster hydration (full descriptions via JSON-LD)

{
  "query": "healthcare",
  "method": "api+html",
  "category": {
    "slug": "healthcare-ai",
    "name": "Healthcare AI",
    "url": "https://artificialintelligencecompanies.com/cat/healthcare-ai/",
    "number_of_items": 8
  },
  "companies": [
    {
      "name": "Abridge",
      "url": "https://artificialintelligencecompanies.com/co/abridge/",
      "description": "Abridge uses generative AI to automatically capture and summarize doctor-patient conversations into structured clinical notes. The company became a unicorn in 2025 with 30% market share in the ambient scribing category, which generated $600M in 2025 (+2.4x YoY). Abridge serves major healthcare systems including Johns Hopkins Medicine, Kaiser Permanente, Mayo Clinic, and Duke Health, with reported outcomes including 78% reduction in cognitive load and 86% of clinicians reporting less after-hours work. Best in KLAS 2025 - Ambient AI Market Leader.",
      "description_truncated": false,
      "source": "json-ld:ItemList"
    }
  ]
}

Shape C — no matches

{
  "query": "quantum-cryogenics-as-a-service",
  "method": "api",
  "endpoint": "/api/search/?q=quantum-cryogenics-as-a-service",
  "matched_categories": [],
  "companies": [],
  "result_count": 0,
  "result_capped_at_5": false
}

Shape D — taxonomy enumeration (used as a precursor to Shape B)

{
  "method": "api",
  "endpoint": "/api/categories/?format=json",
  "total": 12,
  "categories": [
    { "id": 2,  "slug": "ai-automation-agencies",     "name": "AI Automation Agencies" },
    { "id": 10, "slug": "ai-consulting",              "name": "AI Consulting" },
    { "id": 61, "slug": "ai-customer-service",        "name": "AI Customer Service" },
    { "id": 64, "slug": "ai-image-generators",        "name": "AI Image Generators" },
    { "id": 67, "slug": "ai-platforms",               "name": "AI Platforms" },
    { "id": 68, "slug": "ai-receptionist",            "name": "AI Receptionist" },
    { "id": 63, "slug": "computer-vision",            "name": "Computer Vision" },
    { "id": 59, "slug": "foundation-model-providers", "name": "Foundation Model Providers" },
    { "id": 65, "slug": "healthcare-ai",              "name": "Healthcare AI" },
    { "id": 66, "slug": "legal-ai",                   "name": "Legal AI" },
    { "id": 62, "slug": "machine-learning-platforms", "name": "Machine Learning Platforms" },
    { "id": 1,  "slug": "training-data",              "name": "Training Data" }
  ]
}