Most businesses do not need another chatbot that jumps straight to “you should run ads” or “you need SEO.”
They need an assistant that can slow down, understand the business first, ask sensible questions, identify the real marketing problem, and recommend a logical next step. That is the idea behind the downloadable Digital Marketing Discovery & SEO Intelligence Agent framework.
It is a detailed Markdown file that gives an AI a job description, a conversation flow, marketing knowledge areas, qualification rules, and a clear handoff format. It is designed for agencies, consultants, in-house marketers, and business owners who want a more consistent first layer of marketing discovery.
Download the complete Digital Marketing Agent Markdown file
What this AI agent framework actually does
The file is not a magic “do my marketing” button, and it should not be positioned that way. On its own, it cannot see Analytics, Search Console, ad accounts, a CRM, or a website’s live technical condition unless you give the platform approved access to those tools and data.
What it does provide is a disciplined way for an AI to think and respond.
Instead of giving generic recommendations immediately, the agent is instructed to:
• welcome visitors in a natural, non-pushy way;
• learn about the business, goals, current marketing, and pain points;
• recognize the difference between a casual researcher and a sales-ready lead;
• delay technical analysis until it understands the business context;
• discuss SEO, Google Ads, Meta Ads, brand awareness, social media, YouTube, and AI search visibility in a structured way;
• avoid inventing rankings, traffic, conversion rates, or competitor data;
• prepare a clean lead summary for a CRM or sales team when a human handoff makes sense.
That last point matters. A good marketing assistant should know when to keep educating and when to stop pretending it can replace a strategist, account manager, or sales call.
Why the framework is written in Markdown
Markdown is a practical format for agent instructions. It is readable without special software, easy to edit, and works well with platforms that accept pasted instructions, uploaded knowledge files, or file-based agent definitions.
The headings in this file are doing useful work. They separate the agent’s identity, discovery process, qualification logic, advertising frameworks, SEO guidance, brand-awareness guidance, and CRM output. This makes the instructions easier for a person to maintain and easier for an AI to follow than one large, unstructured paragraph.
It is also portable. You can use the same core framework in ChatGPT, Gemini, Antigravity, or another AI workspace, then adapt only the small parts that are specific to that platform.
The important parts of the framework
1. A clear role and tone
The agent is positioned as a consultative digital marketing strategist—not a scripted bot and not a hard-selling salesperson. It is told to be curious, approachable, concise, and strategic.
That is more than a style preference. Tone affects the quality of the information people share. A visitor who feels interrogated is less likely to explain that their real problem is poor lead quality, a broken follow-up process, or a website that does not convert.
2. Discovery before diagnosis
The workflow deliberately puts conversational discovery before website crawling, page-speed checks, competitor research, or recommendations.
This prevents a common AI failure: treating every marketing issue as the same issue. For example, a drop in SEO traffic after a redesign could be caused by migration errors, lost redirects, content changes, tracking problems, seasonal demand, or something else entirely. The right first response is a useful question, not a confident guess.
3. Lead qualification without a robotic script
The file includes simple lead states: visitor, engaged, qualified, hot lead, sales-ready, and cold prospect. These are not meant to pressure people. They help the agent adjust the conversation.
Someone researching SEO for a school project needs a helpful explanation. Someone who has an active ad budget, a decision-maker involved, and a need to grow within 90 days may need a more focused next step, such as a discovery call or audit review.
4. Marketing knowledge that connects to business goals
The framework covers the major areas many agencies discuss:
• technical SEO, content, internal linking, competitor research, and reporting;
• Google Ads, Performance Max, Display, and YouTube;
• Meta Ads for awareness, lead generation, and ecommerce;
• social content, YouTube, brand positioning, and creator collaborations;
• AI-search visibility, answer-engine optimization, and entity consistency.
The useful part is not the list of channels. It is the instruction to tie recommendations back to business stage, audience behavior, funnel maturity, conversion setup, and profitability. A business with no conversion tracking should not receive the same plan as a company with a mature CRM and reliable revenue data.
5. Honest limits on data and audits
This is one of the strongest safeguards in the guide. The agent must not fabricate rankings, traffic estimates, authority scores, competitor metrics, or conversion data.
Without approved access to the right sources, the agent should say what it can and cannot confirm. It can explain likely causes, identify questions to investigate, and suggest what data to review next. It should not present assumptions as an audit.
6. A usable CRM handoff
The included JSON schema converts a useful conversation into structured information: company, contact, website, goals, timeline, budget range, pain points, marketing channels, issues found, recommended services, and notes.
This is the bridge between a helpful chat and a sales process. At first, the agent can create the summary for human review. Later, with a properly secured integration, the same structure can be mapped into a CRM such as HubSpot, Pipedrive, Salesforce, or a custom system.
Before you install it: make it yours
Do not paste the file into an AI platform unchanged and assume it represents your agency. Spend a few minutes adapting it first.
Replace or add:
• your agency name and the type of clients you serve;
• your service area, industries, and minimum project fit;
• the services you actually sell and the ones you do not offer;
• your definition of a qualified lead;
• your preferred handoff action: booking link, email, call request, or CRM task;
• your privacy rules and what the agent must never request;
• approved data sources and the point where a human must review the recommendation.
If you are a local agency, this might mean adding Google Business Profile, local SEO, service-area pages, reviews, citations, and call tracking. If you work with ecommerce brands, you may want more emphasis on feed quality, contribution margin, repeat purchase rate, and creative testing.
Keep the agent focused. It is better to have a strong discovery assistant than a bloated “expert in everything” that gives shallow advice.
How to implement the framework in ChatGPT
ChatGPT has two practical routes: a dedicated GPT where your workspace allows it, or a Project for an ongoing internal workflow.
Option A: Create a dedicated GPT in a supported workspace
1. On the web, open Explore GPTs and select Create.
2. Give it a plain name, such as Digital Marketing Discovery Agent.
3. Add a short description explaining that it qualifies marketing needs, recommends next steps, and creates an internal lead summary.
4. Copy the core behavior from the downloadable Markdown file into the GPT’s Instructions field. The role, tone, workflow, guardrails, and handoff criteria belong here.
5. Upload supporting reference material separately as Knowledge. Good examples are your service descriptions, case studies, pricing guidance, onboarding process, ideal-client profile, and approved FAQs.
6. Enable web search only if you want the agent to research public information. Do not let it describe a live audit as completed unless it actually used the tool and can explain the evidence.
7. Add realistic conversation starters, then test the agent before sharing it.
Separate instructions from knowledge. Instructions tell the AI how to behave. Knowledge files give it facts and source material to use. Mixing everything into an uploaded document often makes behavior less reliable.
At the time of writing, new custom GPT creation is limited to eligible ChatGPT Business, Enterprise, and Edu workspaces. If your account does not have that option, create a Project instead: upload the Markdown file, place the essential behavior in the Project instructions, and use that Project for all marketing-discovery chats. OpenAI is also moving longer-lived custom-GPT workflows toward its plugin-based approach, so keep this Markdown file as your maintained source of truth.
How to implement the framework in Gemini
In Gemini, the closest equivalent is a custom Gem.
1. Open Gemini in a web browser, go to Gems, and select New Gem.
2. Name it clearly, for example Marketing Discovery & SEO Intelligence.
3. Paste the agent’s role, tone, workflow, guardrails, and qualification logic into the Gem’s instructions.
4. Under Knowledge, upload the downloadable Markdown file and any relevant agency documents. You can also add a Google Drive version of a source file when you want later edits to be reflected in the Gem.
5. Preview it using real examples before you rely on it with prospects or staff.
Gem creation and editing happen in Gemini’s web app, although Gems can be used in other supported Gemini surfaces. As with ChatGPT, use the instruction box for operating rules and the knowledge area for supporting documents.
How to implement the framework in Google Antigravity
Antigravity is especially suited to this type of file because it uses Markdown-based custom agents. Unlike a chat-only configuration, you keep the agent inside a project workspace where its instructions can be versioned alongside integrations, forms, CRM mappings, or a future website chat application.
Create this file in your project:
.agents/agents/digital-marketing-intelligence.md
Start it with a small YAML header, then paste the downloaded framework under it:
—
name: digital-marketing-intelligence
description: Discovers marketing needs, qualifies leads, and prepares evidence-aware SEO and paid-media recommendations.
mainAgent: true
subagent: true
—
# Core Instructions
[Paste the complete downloaded Digital Marketing Agent framework here.]
Open Antigravity’s agent selector or the /agents panel and choose the new custom agent. Antigravity can also use it as a specialist called by a larger project agent when you enable it as a subagent.
Be conservative with tools and permissions. The downloaded file is a strategy and discovery framework, not permission to run commands, access a CRM, send messages, or change ad accounts. Add those integrations only when they are needed, correctly authenticated, and subject to an appropriate approval step.
Using the framework in other AI tools
Most platforms use one of the same three patterns:
1. System instructions or agent instructions: paste the behavioral sections here.
2. Knowledge base or uploaded files: attach the full Markdown file plus agency documentation.
3. Tools and integrations: connect only the systems the agent genuinely needs, such as website crawl tools, Search Console, ad reporting, a CRM, or a calendar.
If the platform has a short instruction limit, do not paste every section blindly. Keep the identity, discovery order, lead rules, honesty safeguards, and CRM schema in the instruction field. Store the longer paid-media and brand-awareness frameworks as reference knowledge.
Test the agent before using it with real leads
An agent should be tested against realistic scenarios, not just a friendly “hello.”
Try prompts like:
• “I run a roofing company. Google Ads brings leads, but most of them are not serious. Where should we start?”
• “Our traffic fell after the website redesign. Can you audit it?”
• “We need more online sales before the holiday season. We currently run Meta Ads and email campaigns.”
• “Create a CRM-ready summary of our discussion and tell me what information is still missing.”
In a good result, the agent asks one or two useful questions, avoids unsupported claims, explains the next sensible step, and captures context without turning the conversation into a form.
In a poor result, it recommends every service, invents metrics, asks ten questions at once, or pushes for a meeting before it understands the problem. If you see that behavior, simplify the instructions and add one or two examples of the response style you want.
The difference between a useful agent and an automated sales promise
The first version of this agent should be treated as an intelligent discovery layer. It can make conversations more consistent, help junior staff prepare better notes, surface important gaps, and reduce the time required to turn an inquiry into a qualified opportunity.
It should not be given unchecked authority over advertising budgets, client communications, CRM updates, or marketing claims. Those are implementation decisions that require real data, access controls, testing, and human accountability.
Used responsibly, this framework gives you a reusable starting point for building a marketing assistant that sounds more thoughtful, makes fewer empty claims, and moves the right conversations forward.
Download the complete Digital Marketing Discovery & SEO Intelligence Agent framework and adapt it to your business before deploying it.
