SPD says the communications director only used AI tools a handful of times, and only “to evaluate their utility”
By Erica C. Barnett
The Seattle Police Department communications director, Barbara DeLollis, used AI chatbots that are not approved for city use to compose a sample script for a woman preparing for her first media interviews, to produce a list of “Interesting best practice on-camera ideas for big police Department,” to produce a “Comprehensive Communications Toolkit for a Police Department Exiting a Consent Decree,” and to to rewrite a published blog post about a nuisance motel on Aurora.
That prompt reads, in full, “hi make this a better story for the public of a city that doenst liek crime or disorder” (sic).
DeLollis used ChatGPT to produce the sample blog post, and Perplexity to produce the other documents, according to records PubliCola obtained through a public disclosure request. The city’s information technology department confirmed that neither program is approved for use by city employees.
Last September, after noticing that a number of the department’s public communications had many of the hallmarks of AI, PubliCola requested “documents detailing all uses of generative AI” for the first nine months of the year by communications staff as well as Police Chief Shon Barnes and his staff.
The Seattle Police Department provided seven documents, all produced by DeLollis, and closed our request. We asked SPD to confirm that they are asserting that Barnes has never used generative AI, and that the seven documents represent every single use of AI by DeLollis and SPD’s entire communications team. They said yes.
However, the records themselves include two AI-generated documents for which SPD did not produce the written prompts that preceded them—an obvious omission of records responsive to our request that raises concerns about whether the documents really represent every use of AI by DeLollis or other staffers.
Last year, an anonymous person filed two complaints ahout SPD’s use of AI with the Office of Police Accountability, citing the apparent use of AI in a bio of Barnes’ chief of staff, Alan Ricketts, a bullet-pointed statement from Barnes about a violence prevention and enforcement effort, and other documents. The evidence in those complaints included a blog post full of passive-voice, AI slop-style sentences such as “On Thursday, we were confronted with a targeted homicide occurring in front of a place of worship.” That complaint resulted in a supervisor action (essentially, a reprimand).
An SPD spokesperson responded to PubliCola’s nine detailed questions with a statement that read in part: “Last year, a department employee tested various AI tools to evaluate their utility in communication functions like editing, interview preparation, and blog strategy to see if they could offer fresh perspectives.”
The Office of Police Accountability “determined that using AI tools in this way without appropriate acknowledgement was a violation of city policy at that time,” the spokesperson said. “The department does not condone using generative AI to write narratives or communications.”
PubliCola is supported entirely by readers like you.
CLICK BELOW to become a one-time or monthly contributor.
A spokesperson for Mayor Katie Wilson told PubliCola, “Unapproved AI software is now blocked on city devices to ensure compliance with critical privacy, transparency, and records protections. The Mayor expects employees to use these tools in compliance with city policy.”
DeLollis’ prompts are riddled with typos that make them challenging to credit as official SPD work product and sometimes hard to interpret.
For example, in one Perplexity prompt—”Lost thengoala for a daily police blog where uoupost both police actions and responses to 911 callbut also show in compelling ways the other impactand resorts of work that a cutting edge evidencebased department does”—DeLollis appears to be asking the AI to define the goals for SPD’s Blotter blog. (The response, which includes generic advice like “Combining transparency with compelling storytelling and data will build trust while showcasing the full scope and positive impact of a modern police force,” seems less than useful.)
In two other conversations with the Perplexity chatbot, DeLollis appears to be seeking advice for a female employee doing her first media interviews and who, as a woman, tends to overprepare for things. SPD did not respond to our questions about the purpose of these prompts or whether they were on behalf of a specific woman.
“So we know why woken over prepare for media interviews but for our client we want to frame this advice on a positive way to prevent them from feeling negative. Help,” one of these prompts reads. “Frame this in positive way for client who is going to need prep for her first media interviews. Women typically over prepare for research drive reasons. It is t helpful though,” another begins.
The records SPD provided for the latter Perplexity query include an ongoing conversation between DeLollis and the chatbot, including a request for a sample script and two requests for academic research.
Because it’s AI (and AI sucks), Perplexity responded to the prompt about helping a woman avoid over-preparing with a list of reasons why it’s important to prepare. In the second conversation, the chatbot added 23 “sources” that included 10 duplicative links and several posts that were unrelated to the question, including guides for interviewers about talking to women who are researchers or subject-matter experts.
Perplexity also produced two guides for communicating about the end of the federal consent decree. (These are the two documents for which SPD did not include the AI prompts). The first is a series of bullet-pointed lists; the second, mentioned earlier, is more of a media “kit,” with sample op/ed language and social media posts, like this suggestion for a post on X: “We’ve made big changes in how we train, respond, and build trust. Now that we’ve met the federal standards for reform, our work continues—with you.”
In the final prompt, DeLollis asks Perplexity to come up with “Interesting best practice on-camera ideas for big police Department.”
It’s unclear whether DeLollis created the chat prompts from a city of Seattle computer or personal device. Washington’s public disclosure law requires city employees to produce all records that are responsive to a request, including those produced on personal devices or using personal emails or cell phone numbers.
According to the central staff analysis, Mayor Bruce Harrell’s proposed budget is unsustainable and relies heavily on fiscal sleight-of-hand to come up with a balanced budget in 2026, tumbling precipitously into massive deficits in 2027 and beyond. These tricks include relying on a one-time $141 million fund balance left over from 2025, which won’t be there to balance the budget next year; funding programs that will be necessary long-term, like food assistance for people losing federal benefits, with one-time resources, so that they don’t count toward future deficits; and assuming a $10 million “underspend” every year in the future, allowing the mayor’s budget team to chop $10 million off each year’s expenditures automatically without actually making cuts.
By Erica C. Barnett