Skip to content
Menu
A green valley with a river threading down the middle

In ChatGPT and AI assistants

ChatGPT knows the world. Agentcy knows
your clients.

Your team already has an assistant open all day. The Agentcy MCP App gives it your clients’ own record to work from, so nobody has to learn another screen.

See it in your workspace
↓

One question, followed through.

A newsjack request in ChatGPT with Agentcy connected, from the question to a pitch waiting for your click.

↓

The same question, a different answer.

What changes when the assistant has Agentcy behind it.

“What’s in the news for this client?”

ChatGPT on its ownA web search, with no sense of what matters to the client

With AgentcyMore than 300 monitored outlets and government feeds, scored for relevance to this client

“Who should we pitch?”

ChatGPT on its ownNames from the open web, often out of date

With AgentcyActive journalists, linked to what they wrote and to your past pitches

“What has our spokesperson said?”

ChatGPT on its ownOnly what it can find in your files

With AgentcyEvery quote from coverage, tracked for each spokesperson

“How did coverage go last month?”

ChatGPT on its ownNo record of coverage

With AgentcyEvery article, with reach, tier and sentiment against the month before

“Comment on this story for us”

ChatGPT on its ownA draft in a chat window

With AgentcyWritten in house style, approved, sent by a person, then tracked

“What do I need to know today?”

ChatGPT on its ownAnswers only when asked

With AgentcyComment opportunities, enquiries and competitor moves, every morning

↓

Why it’s better

Same model. Better context.

Give an assistant a breaking story and ask for a comment, and it writes something fluent and general, with a blank where the spokesperson should be. A journalist can’t quote “[Spokesperson Name]”.

With Agentcy connected, the same model has four things to work from.

01

It knows the spokespeople

Each client’s approved spokespeople and their subjects, so the comment comes from the right person by name.

02

It read the actual story

The article comes from Agentcy’s news feed, already scored as relevant, so the comment deals in specifics.

03

It knows the positioning

The client’s messaging and what it sells, so the comment argues from what the client actually does.

04

It writes in house style

The client’s style guide and ground rules apply to every draft, banned words included.

↓
Stepping stones crossing a shallow upland stream

No new interface
to learn.

Agentcy’s boards open inside the conversation
↓

What people ask it.

Three questions to try first. Each opens the working board inside the chat, so you can act on the answer there.

All workflows
“What needs me today?”
In ChatGPT, every morning

Start the day across every client

  1. 01Reads what changedNew coverage, comment ideas, journalist enquiries, approvals and overdue tasks
  2. 02Ranks what mattersThe ten items to act on first, each with a reason
  3. 03Opens the workEach item opens its board, inside the assistant
  4. 04Arrives every morningRuns as a scheduled task, so nobody has to ask

An assistant aloneknows nothing happened overnight unless someone pastes it in.

“Prep our CEO for tomorrow’s interview”
In ChatGPT or the web app

Brief a spokesperson for an interview

  1. 01Profiles the journalistTheir recent articles and past coverage of the client
  2. 02Recalls the client lineKey messages, proof points and ground rules
  3. 03Checks the recordWhat the spokesperson has said before, to stay consistent
  4. 04Writes the briefA one-page briefing, saved to the client’s documents

An assistant alonegives generic interview tips, with no record of either person.

“How did coverage go for the client last month?”
In ChatGPT, the web app or your own agent

Report on the month’s coverage

  1. 01Gathers every pieceLogged as it appears, with reach, tier and sentiment
  2. 02Measures against goalsProgress on the client’s KPIs and on the month before
  3. 03Adds the wider pictureShare of voice, spokesperson quotes and AI visibility
  4. 04Ready to sendA gallery and a report, emailed to you to forward

An assistant alonehas no record of coverage, so someone compiles it first.

↓

Your controls stay in charge.

Connecting an assistant doesn’t loosen anything. It reaches Agentcy through the same permissions as a person.

01

Sign in through Agentcy

OAuth 2.1. The assistant holds a token, never a password, and access ends when an account is switched off.

02

Permissions on every call

Each request is checked against that person’s access today. The model can’t reach a client they can’t see.

03

Switched on by your admins

AI app connections are on or off for each account, in Settings.

04

Only a person can send

The assistant can draft a pitch. Only a person clicking Send in the Agentcy panel sends it.

05

Logged and reversible

Every call is recorded: who, what and when. Changes can be undone.

06

Outside text is data

Article and email text is never treated as an instruction to follow.

↓

Start where you already work

How do you want to work?