Three years into the generative AI cycle, the early adoption hype has settled into something more useful: a clearer picture of which tools are doing real work in agency operations, which are still in pilot, and which were oversold and quietly shelved. A scan of agency operations leaders across mid-sized and large independent firms produces a consistent picture of what is being deployed and what it is doing.
Tools that have stuck
Media database AI features. Muck Rack and Cision have both rolled out generative features integrated with their journalist databases — pitch drafting against a specific reporter's beat history, predictive matching of stories to outlets, automated summarization of incoming press hits. Adoption is wide. The features save real time on tasks that used to require hours of manual work.
General-purpose LLMs for drafting. Most agencies use ChatGPT, Claude, or Gemini for first-pass drafts of internal documents, client briefs, social copy, and meeting notes. The work product is universally edited before going to clients, but the time savings on first drafts is meaningful. Estimates from Cision's State of the Media and Muck Rack's State of Journalism reports suggest most practitioners save several hours a week.
Transcription and meeting summarization. Otter, Read.ai, and similar tools have become near-ubiquitous in agency workflow. The output quality is good enough that meeting notes are no longer a manual task in most operations.
Custom GPTs and Claude projects for repeatable workflows. Agencies have built internal tools using OpenAI's custom GPTs or Anthropic's Claude projects for tasks like brief-to-pitch translation, boilerplate generation, and competitive media monitoring. These are usually agency-specific and not commercially licensed.
Tools in pilot
AI visibility audit platforms. Several vendors have launched products that promise to measure brand presence across LLMs. The category is young, the methodologies vary, and most agencies are evaluating multiple options rather than standardizing on one. The honest read is that this category will consolidate within 12 to 18 months.
Automated earned media impact scoring. Several vendors are pitching AI-powered approaches to media impact scoring beyond traditional advertising-equivalency or impression metrics. Adoption is uneven; the tools work better for some categories than others.
AI image generation for owned content. Some agencies use Midjourney, DALL-E, or similar for owned content visuals. The legal posture is still settling, and most clients prefer human-created imagery for anything with even modest visibility.
Tools that did not stick
AI-driven press release generators. Several vendors pitched fully automated release generation in 2023 and 2024. The output quality was poor, the legal review burden offset the time savings, and most agencies have abandoned the category.
Synthetic spokesperson video generation. Briefly trendy, mostly retired. Audience trust dynamics around synthetic video are unfavorable for nearly all use cases that brands actually want.
LLM-powered influencer matching. Promised better results than existing platforms, mostly delivered worse ones. Influencer evaluation turns out to require qualitative human judgment that current tools do not replicate well.
What agencies are spending
The honest economics: most mid-sized and larger agencies are spending in the low-to-mid five figures per year on AI tooling beyond their existing media database subscriptions. The spend is split across general-purpose LLM access (often ChatGPT Enterprise or Claude Team), specialized tools, and internal tooling development.
The spend is rising but not exploding. The capability gains have outpaced the cost increases for most agencies, which means tooling investment is generating positive ROI on internal operations.
What this means for staffing
The structural question every agency operations leader is now grappling with is what happens to junior roles. The early hype suggested wholesale displacement of associate-level work. The actual pattern is more nuanced.
Junior staff are using AI tools heavily, and their per-capita output has risen. The work that used to be entry-level — pulling clip reports, drafting status documents, building media lists — is faster. But the time freed up is mostly being absorbed by other work, not used to reduce headcount.
Where headcount has shifted, it has been from larger junior pools to smaller pools of better-trained associates supported by AI tooling. The agencies seeing the most operational lift are the ones investing in junior training on tool use rather than treating AI as a substitute for hiring.
A working framework for evaluation
Agencies considering new tools should ask three questions. Does it solve a problem that takes meaningful agency time today? Is the output quality good enough that the time savings are real, or does the editing burden offset the gain? Is the tool from a vendor that is likely to be operating in 12 months?
The third question matters more than it sounds. Several heavily-pitched AI tools from 2023 and 2024 are no longer in market. Agencies that standardized on them spent transition cost. The category is consolidating, and conservative selection beats early adoption for most operations.
