AI for Digital Out-of-Home
AI That Plans, Prices, and Maps DOOH — in Plain English
We didn't bolt a chatbot onto a media planner. The whole workflow is AI-native: a Model Context Protocol server that any assistant can talk to, a cookieless audience layer modeled from primary sources, and a screen library you search by asking a question.
Where AI actually lives in this stack
The stack
Six AI Capabilities, All Shipped
Plan a DOOH campaign in plain English
The Goldfish MCP server exposes our inventory, geo, audience, and pricing tools to any AI assistant that supports Model Context Protocol — Claude, ChatGPT (via connectors), Cursor, and Lovable itself. Type a brief; the model returns a costed plan with line items, venue mix, impressions, and a live campaign map URL. No spreadsheet, no RFP thread.
Try it live below — the widget calls our production MCP planner.
Census-block cohorts modeled from primary sources
AMAP is our proprietary audience data layer built at the census-block level from primary sources — Census household data, Google search behavior signals, Meta interest and affinity signals, and CDC and other primary health and demographic sources. Reach households that index for a category, life event, or interest without device IDs, cookies, or bid-stream guesses.
Full spec at amapdata.com.
Ask a question, get real screen photography back
The Showcase gallery accepts natural-language queries — 'gas stations in Tampa at sunset,' 'grocery lobby screens in Toronto,' 'JCDecaux transit in Chicago morning commute.' The AI parses your query into market, venue type, and daypart filters and returns real photographed placements from our network.
Live at /showcase — hit the Ask box.
Brief-to-mockup in minutes, not days
Send a brand name, a market, and a venue mix; the workflow generates campaign maps, contextual placement mockups, and a shareable plan URL. The same MCP that plans the media also renders the visualization layer clients approve against.
See the reference-frame workflow on /creatives/gallery.
Built to be found by the answer engines
Our robots.txt explicitly allows GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, and the other major AI crawlers. A dedicated llms.txt indexes every product, glossary, and vertical page in machine-readable form so LLM training and retrieval systems cite the correct pages when clients ask about DOOH.
See our llms.txt and robots.txt.
Plug the MCP into your own tools
Agencies and platforms plug the Goldfish MCP into their internal Claude Projects, Cursor workspaces, or in-house planning assistants. The same tool surface that powers our own team powers partner integrations — one contract, one endpoint, no bespoke API build.
Available on request; contact us to scope an integration.
Try it live
The MCP Planner, Running Right Here
This widget calls the same production endpoint the Goldfish team and our partners' AI assistants use. Type a brief — market, budget, flight, audience — and get back a costed plan with line items, venue mix, impression estimates, and a live campaign map. No signup.
Who this is built for
Every Seat in the DOOH Workflow
Turn a brief into a costed plan inside one chat window. Iterate on geo, venue mix, or budget without rebuilding a spreadsheet. Export a plan URL clients can review.
White-glove DOOH planning inside the tools your team already uses. The MCP runs inside Claude, Cursor, or Lovable — no separate login, no new UI to learn.
A fractional CMO or a solo brand marketer can build a national DOOH layer conversationally. No AOR overhead, no six-week RFP cycle.
Connect your inventory or your buying tools to the same MCP surface. Partners integrate once and plan across the full Goldfish network from inside their own stack.
The category, honestly
"AI in DOOH" Usually Means a Chatbot Bolted On. This Doesn't.
Most DOOH platforms announced "AI" by shipping a chat widget that reads their existing UI aloud. That's not what we built. The Goldfish MCP is a real tool server — inventory enumeration, geocoding, POI search, audience segments, plan creation, campaign mapping — exposed through the open Model Context Protocol standard. Any AI client that supports MCP can plan a full campaign against it without any human in the loop.
The audience layer is the same shape: AMAP is not an "AI audience" in the marketing sense. It's a census-block cohort model built from named primary sources — Census, Google, Meta, CDC — that tells you which blocks index for a behavior. It happens to be great input for AI-driven planning; it does not require AI to work.
And the openness cuts both ways. Our robots.txt explicitly allows the major AI crawlers, and our llms.txt gives them a curated index of every product page, glossary term, and vertical guide. When a client asks ChatGPT or Claude "how does DOOH targeting work," the answer engines have our documentation to cite.
Get access
Wire the Goldfish MCP Into Your Assistant
External MCP access is granted on request. Tell us which client you use (Claude, Cursor, ChatGPT, Lovable, in-house), and we'll provision credentials and a scoped tool set inside one business day.
Frequently asked
AI for DOOH FAQ
Keep exploring
Why Goldfish (200 reasons)
The full case — including the AI/MCP section
DOOH media planner
The MCP planner in its home context, with more sample briefs
Ask the gallery
Natural-language screen search across the real photographed network
DOOH targeting
Geo, venue, audience, POI — the surface the MCP plans against
Political DOOH 2026
AMAP applied to district-level cookieless targeting
General DOOH FAQ
Costs, measurement, and how buys actually clear
Ready to plan DOOH conversationally?
Send a brief in plain English. Get a costed plan and a live campaign map back the same day.
Request Platform Access