How ward51.com Gets Made: Tools, Process, and How to Pitch Us
How this was made
This site has a disclosure on every post: researched and drafted with AI assistance. Most publications that carry a line like that don't explain what it means. This one does — and this post is the full explanation.
If you've ever wondered what happens between your suggestion and the published article, or why some briefs produce sharp pieces and others produce something vague and forgettable, this is for you. It covers the full production process, the specific tools the AI assistant has access to, and how to write a suggestion that actually works.
The Production Process, Step by Step
Every article on ward51.com follows the same pipeline, regardless of whether it's a civic data investigation, a Fire FC tactical breakdown, or a Chicago Fire episode recap.
1. The Brief
Everything starts with a question or a direction from the editor. The brief is the intellectual core of the piece — it sets the angle, the argument, the audience, and the tone. A vague brief produces a vague article. The AI assistant is a capable executor; it is not a concept generator. The direction has to come from a person.
A good brief answers: What is the central question? What's the argument, or what should the data reveal? Who is this for? What angle hasn't been done? What primary sources should anchor the piece?
2. Research
Once the brief is clear, the assistant runs research. For Chicago civic topics, that means querying the City of Chicago Data Portal, Cook County data systems, and Illinois state databases for authoritative municipal data. For city council topics, it means pulling real legislation records, aldermanic vote histories, and meeting agendas from the Chicago City Council's official systems. For everything else, it means web searches against reliable sources.
The research phase is where ward51.com's model earns its keep. What would take a reporter several hours — finding the right dataset, querying it against specific parameters, cross-referencing vote records, surfacing the specific ordinance number — takes minutes. The constraint isn't speed. The constraint is knowing which questions to ask.
3. Drafting
The assistant writes a full draft, following the ward51.com writing style guide: data first, opinion second, plain language throughout, no passive-voice evasion of accountability, and no hedged conclusions after 2,000 words of evidence. Internal links to related posts are woven in. Data visualizations are generated as SVGs when the data warrants one.
4. Editorial Review
The editor reads everything before it publishes. The voice, angle, and conclusions are editorial decisions. The AI can execute a thesis; it can't decide whether the thesis is worth making. That judgment is human. Every piece is approved, corrected where needed, and published by a person who's willing to put their name on it.
The brief is displayed publicly alongside every article — not as a caveat, but as a record. You can read exactly what was asked for and judge whether the piece delivered it. That's the model. We've explained the philosophy behind it, and we've walked through a real post's lifecycle in detail.
What Tools the AI Assistant Has Access To
This is the part most AI-assisted publications don't share. Here's a full accounting of the tools available during article production.
Web Research
The assistant can search the web in real time using the Brave Search API. This is used for breaking context, verifying facts, finding primary sources, and researching topics outside Chicago municipal data. Results include titles, URLs, and source snippets — the assistant synthesizes and cites these, it doesn't reproduce them verbatim.
Chicago Open Data Portals
This is the engine behind the civic section. The assistant has direct query access to three Socrata open data portals:
City of Chicago — crime incidents, CTA ridership, 311 service requests, building permits, business licenses, restaurant inspections, city budgets, capital projects, and hundreds of other datasets maintained by the city.
Cook County — property tax assessments, court records, public health data, and county government operations.
State of Illinois — state budgets, education funding, IDOT transportation data, and FOIA request logs.
The workflow: discover relevant datasets by keyword, inspect the schema, then run targeted queries using SoQL — a SQL-like language for filtering, grouping, and aggregating the data. Every dataset query is logged as a citation. If a dataset hasn't been updated recently, that gets flagged as a freshness caveat in the research notes.
Chicago City Council Systems
Chicago civic pieces often require more than aggregate statistics — they require knowing who voted for what, when, and whether they showed up. The assistant has access to:
Legislation search — find ordinances, resolutions, and orders by keyword, date, status, or type. Every piece of legislation is cited by its official file number (e.g., O2026-0042) and linked to the sponsoring alderperson and their ward.
Council member lookup — current alderpersons, their ward boundaries, and their committee assignments.
Vote history — roll call votes for any council member, filterable by date range.
Meeting schedules — upcoming and past meetings for the full City Council and key committees (Finance, Zoning, Budget).
Aldermanic metrics — attendance rates, legislation counts, and substantive versus routine bill breakdowns for any of the 50 wards.
Ingested meeting database — a local store of processed meeting records with full attendance rolls, complete agenda items (600+ per typical full council meeting), contested vote flags, and per-alderperson vote breakdowns. This is what powers detailed reporting like the Q1 2026 City Council Report Card.
Post and Content Management
The assistant can list all existing posts (to check for related content and internal linking opportunities), read full post content, create new drafts, and update existing ones. It cannot publish — that action requires explicit editor approval. Every draft is saved with research notes, editor notes, and the original brief attached.
Tags and Taxonomy
The assistant checks the existing tag library before saving a post and applies relevant existing tags. New tags are only created when no suitable existing tag covers the topic. This keeps the site's taxonomy clean and internally consistent.
Media and Images
The assistant can list existing media, generate SVG images (diagrams, charts, maps, hero illustrations), and attach them to posts. All generated images include descriptive alt text. Hero images are required for every published post. Data visualizations are generated as SVGs whenever a dataset has three or more categories, a time dimension, or a geographic dimension — the threshold for "just put it in a chart" rather than listing numbers in prose.
What the Assistant Cannot Do
A few important constraints worth naming explicitly. The assistant cannot access paywalled sources. It cannot file FOIA requests or retrieve documents that aren't already in a public data portal. It cannot independently verify a claim against an offline or non-digitized record. And it cannot make the editorial judgment calls that determine whether a piece is actually worth writing — that's the editor's job, not a tool's.
How to Write a Suggestion That Works
The suggestion box exists because good questions come from people who live in Chicago, follow Fire FC, or watch the show — not just from the editor's own frustrations. But the quality of a suggestion directly determines the quality of the resulting piece. Here's how to make your suggestion land.
Be specific about the question
"Write about TIF districts" produces a 3,000-word overview that covers everything and reveals nothing. "Write about which TIF districts in Chicago are sitting on the largest unspent balances, and whether they're in neighborhoods that also have the worst infrastructure ratings" produces an investigation with a real finding at the end.
The difference: the first asks for a topic. The second asks for an answer.
Name the angle, not just the subject
For civic pieces: What's the argument? What do you expect the data to show — or what have you noticed that you want someone to dig into? For Fire FC: What's the tactical question, the roster concern, the stat anomaly? For the TV show: What's the take, the character arc, the moment that deserves more than a recap?
Naming the angle isn't mandatory — sometimes "I don't know what I'll find, just look into X" is the right brief for a data investigation. But if you have a hunch or a hypothesis, share it. It sharpens the research.
Point to the primary source if you have it
If you've seen a stat, a news report, a council vote, or a data portal entry that prompted your question — include it. A link to a specific dataset, ordinance number, or news story cuts research time significantly and ensures the piece is anchored to the same source that caught your attention.
Specify the section
ward51.com has three sections with distinct tones and standards. Civic is data-forward and serious. Fire FC is analytical and fan-perspective. Chicago Fire (the show) is fun and opinionated. Knowing which lane you're pitching helps calibrate the voice and length before the first sentence is written.
A template that works
You don't need to use a form. But if you want a format that tends to produce good results, this structure covers the important pieces:
Section: [Civic / Fire FC / Chicago Fire]
Question: [The specific thing you want answered or argued]
Angle: [Your hypothesis, hunch, or the finding you're expecting]
Source (if any): [Link, dataset name, ordinance number, article]
Anything else: [Related posts to link, comparisons you'd find useful, any time sensitivity]
That's five fields. You don't need five paragraphs. A sentence per field is enough.
What Happens After You Submit
The editor reviews suggestions and decides what gets assigned. Not every suggestion becomes a post — some are too narrow, some require sources that aren't publicly available, some duplicate existing coverage. But every suggestion that points at a real question and a real data source gets a serious look.
When a suggestion becomes a piece, the original brief is published alongside it. If your question shaped the investigation, that's part of the record. That's not a courtesy — it's how the model works. The transparency goes all the way down.
If you want to understand why this site was built the way it was, start here. If you want to see what the AI/human collaboration looks like under the hood on a specific piece, read this. Then come back and pitch something.
The data's all there. Someone just has to ask the right question.