tool: scrape_url
Read the source
Drop a landing page, a competitor ad, a press release. The agent fetches the page and pulls value props, audience cues, tone into the brief.
→ pulled 4 sections from yoursite.com
the loop
A chatbot answers a question. An agent finishes the task.
01
Read
Brief in chat or via guided wizard. Drop URLs and the agent scrapes them before writing.
02
Draft
A purpose-built writer drafts copy in the native shape of each platform. Visuals are generated in parallel.
03
Score
A separate predictor grades every emitted campaign 0–100. Hook variants get ranked before they reach you.
04
Iterate
Ask for a sharper hook, a shorter caption, a different angle. The agent rewrites and re-scores against the same brief.
the toolbox
Five tools. One loop.
These are the real tools wired into the agent — not a wishlist. Each one is called when the brief calls for it; none of them is a black box. You can see the trace.
tool: web_search · exa_similar
Anchor in the now
When the brief needs context the agent doesn't have — a trend, a competitor, a current event — it searches the live web before writing.
→ 6 results · 2 similar pages
tool: draft_campaign · ask_draper
Write to platform
Calls a purpose-built writer for ad copy, in the native shape of each platform. Variants and rewrites run on the same writer.
→ meta + tiktok + pinterest drafted
tool: generate_image
Visual in the same loop
On visual platforms the agent commissions an image from the same brief the writer just produced. Art direction stays consistent.
→ 1:1 still · pour-over scene
tool: score_copy
Grade before it ships
Every emitted campaign auto-scores 0–100. Hook variants are ranked. You see the number alongside the prose, not after.
→ 78 / 100 · strong
What the agent doesn't do: publish to ad accounts, manage spend, run optimization on live campaigns, store a persistent brand voice across sessions. You ship.
Pattern, not productChatbot vs. agent
A great chatbot
is not a great marketing agent.
Dimension
Chatbot pattern
Agent pattern
Shape
One response per prompt
Multi-step loop with tool calls between turns
Inputs
Whatever's in the prompt window
Prompt plus live web pages, attached files, scraped URLs
Output check
You read it and decide
Agent checks its own output against the brief before returning
Platform shape
Generic blob, you reformat
Drafts into platform-specific structure directly
Quality signal
Subjective read
Runs a scoring model on its own drafts
The RatesMetered by drafts
Pricing built around drafts , not seats.
Free
A taste of the desk.
$0
3 drafts / month
- ~3 drafts per month
- 7-day conversation history
- Single-user
- Community support
Most chosen
Pro
For the working marketer.
$20/mo
150 drafts / month
- ~150 drafts per month
- Full conversation history
- Priority model access
- Email support
Pro+
Small agencies & heavy use.
$49.99/mo
500 drafts / month
- ~500 drafts per month
- Full conversation history
- Priority model access
- Priority email support
Enterprise
For brand and agency teams.
Let's talk
Custom draft allocation
- Custom draft allocation
- Dedicated onboarding
- SSO & audit logs
- Direct line to the team
Every plan is metered by drafts. No overage bills, no surprise charges — if you hit the cap, you upgrade or wait until the period resets.
Frequently asked
The agent questions we keep getting.
What is an AI marketing agent?
A language model that runs a marketing task as a multi-step loop, calling outside tools (research, scraping, image generation, scoring) between steps. A chatbot returns one answer; an agent completes a task, including its own intermediate steps.
What is agentic marketing?
The workflow shift away from prompting a generalist chatbot per draft and toward handing an agent a brief, letting it do the research, drafting, formatting, and self-scoring, and reviewing the finished work yourself. The marketer still owns final-ship decisions.
Is ChatGPT an AI marketing agent?
ChatGPT can be used in an agentic pattern when wrapped with tools (browser, retrieval, file I/O). On its own, in a default chat session, it follows the chatbot pattern — single response per prompt, no self-iteration. Whether something is an agent depends on the wrapping, not the underlying model.
Can Claude do marketing?
Claude is a general-purpose model and can write marketing copy. Whether it acts as a marketing agent depends on the surrounding software — what tools it can call, whether it self-checks, whether it can iterate within a turn. Draper's orchestrator runs on a general-purpose LLM and our purpose-built writer is the model that produces the ad copy itself.
What is the difference between AI marketing automation and an AI marketing agent?
Automation runs predefined rules — “if X then Y”, scheduled sends, audience triggers. An agent makes step-level decisions inside a task based on what the brief and the intermediate results call for. Automation is a flowchart; an agent is a loop that picks its next step. Most current tools are one or the other; some claim both.
Is Draper fully autonomous?
No. Draper completes drafting tasks — research, copy, image, score — and stops. It does not publish ads, set budgets, monitor live performance, or make spend decisions. The hand-off is to you.
Which tools does the Draper agent actually use?
scrape_url for pulling content from URLs; web_search and exa_similar for live research; attach_file for files you upload; draft_campaign and ask_draper for the writer model; generate_image for visuals on visual platforms; score_copy and auto-scoring on emit for the quality check. The list is in the shipping code, not roadmap.
Should I use Claude for marketing instead?
If you need a generalist for ad-hoc tasks, Claude is excellent. If you specifically want ad copy in the shape of paid ads, with a built-in scoring step, that is what Draper's purpose-built writer is for. The two are not mutually exclusive — Draper's orchestrator is itself a general-purpose LLM, and the specialist writer sits inside the loop as a tool.
The final wordWrite the ads
Write the ads
you wish you'd written.
Free to start. Metered by drafts, not seats.
no credit card · sign in with Google