Policy think tanks usually publish hundreds of reports, working papers, briefs, and op-eds over years, and the resulting archive becomes both the institution's core asset and its worst interface. A journalist on deadline wants to know what the think tank has published recently on energy transition financing in emerging markets, and a static search box returns 87 results sorted by date. SleekAI reads the publication archive from WordPress and the full report PDFs from OpenAI Files (up to 1 GB per file), then answers "what's your recent work on X" with three named publications, the lead scholar, and the core finding from each.

Scholar discovery is the other recurring task. Reporters, government staffers, and academic peers want to know which scholar at the institution covers a specific topic, and they want a fast answer with the scholar's bio, current research focus, and contact path. Storing scholars as a custom post type with focus areas in postmeta makes the bot a working scholar-directory query: "who covers semiconductor industrial policy" returns the right name, a one-paragraph bio, and a link to recent publications, with the press contact for interviews referenced separately.

Event and program awareness is the third surface. Think tanks run a calendar of public events, webinars, and convenings that lives in a custom post type with date, format (in person versus virtual), and registration link. A visitor asking "are you running anything on tariffs this month" gets the actual event with date, format, and link. Display conditions can hide the bot on board-restricted research pages and surface it on public landing pages where it matters most. Conversation logs in WordPress reveal which research areas attract the most reader interest, which helps the publications team understand what to push.

Policy think tanks accumulate a publication archive that is the institution's most valuable asset and its least usable surface. A typical institute publishes hundreds of working papers, briefs, op-eds, and event recordings over a decade, and that body of work is functionally invisible to a journalist on a 4 PM deadline who needs to know what the institute has said about a topic in the last six months. The institute's own comms team usually carries the search burden, fielding the same kinds of questions about recent work, the right scholar to talk to, and upcoming events, while the underlying content sits structured and ready to be queried.

A semantic chatbot turns the archive into a real interface. The bot reads publications, scholar bios, and events from WordPress and the full PDFs from OpenAI Files, then answers "what's your recent work on X" with cited names and findings rather than a search-result page. The result is not a replacement for the comms team, it is a force multiplier for the team's relationship-building work.

Attribution discipline is the other half of the value, and one think tanks must get right. Most institutes publish through named scholars rather than as a single institutional voice, and the bot's system prompt has to honor that. "Per the February 2026 working paper by Lin and Park" is the right kind of cited attribution, not "the institute believes." Done well, the bot makes the institute's work more accessible without flattening it into an editorial line.

Privacy and access posture matter for institutes with board members, donors, and confidential research. Multibot and user-role display conditions let separate scoped bots exist for board members, press, and the public without any cross-contamination at the data-source level. The board bot reads board-only content; the public bot reads public publications only; the press bot has its own scope with an awareness of embargoes when those exist.

Conversation logs sit in WordPress, which matters when your work happens to touch sensitive jurisdictions and you want full custody of who asked what.