AI tools do not fail because prompts are weak. They fail because customer context is weak.
That matters now, because the market is moving toward connected systems. Teams with mature CRMs are wiring tools like Claude into their customer data through MCP, the Model Context Protocol, so the model can work with real accounts, notes, and history instead of a blank page. In early-stage companies, though, the more common reality is simpler, and messier. Your best context lives in DMs, call notes, inboxes, docs, and memory.
The answer is not to wait for a proper CRM. The answer is to build a minimum viable context layer, then feed it to AI in a disciplined way.
The shift is real
Recent operator chatter has made the shift plain. Emily Kramer noted that 33 percent of paid MKT1 subscribers were already using the MKT1 MCP server, and the MKT1 MCP Showcase is set to surface more connector examples from tools like Attio, Airtable, Framer, Mutiny, Profound, and Softr. The message is clear, customer context is becoming a first-class input to AI marketing, not a nice-to-have.
For funded teams, the obvious move is to connect the CRM. For early-stage teams, the more important move is to stop pretending that context does not matter until the CRM is “real.” It matters first.
Prompts are not enough
Generic AI output usually comes from one of three problems, the prompt is vague, the source material is thin, or the model has no access to actual buyer language.
Most teams overestimate the prompt and underestimate the data. They ask Claude or ChatGPT to write a homepage, a nurture sequence, or a LinkedIn post, then wonder why the output sounds interchangeable. The model is doing what it can with what it has, which is often little more than a category, a value prop, and a tone request.
If you want better output, you need better inputs. Not more words, better evidence.
AI is only as useful as the customer context behind it.
MCP, simply put
MCP, or Model Context Protocol, is a standard that lets AI tools connect to external systems in a more structured way. In practice, that means a model can pull from connected sources, rather than relying only on what a human pasted into the chat.
For marketing teams, that can mean CRM records, call transcripts, event lists, account notes, or product usage data. The point is not the acronym. The point is that the model can see the customer world with less manual copying and pasting.
That is powerful for a team with Attio or HubSpot in place. It is still relevant for a team with a spreadsheet and a folder of call notes. The structure matters more than the system.
What mature teams do
The funded-team version of this pattern is straightforward. They use the CRM as the source of truth, then expose useful objects to AI.
Common workflows include:
- turning unstructured call quotes into recurring themes,
- pulling closed-lost reasons into objection clusters,
- generating follow-up emails from meeting history,
- building event or account lists for outbound and nurture,
- drafting pages and ads from real buyer language, not internal jargon.
Attio has been one of the visible examples in this conversation, because its structure makes it easier to work with customer data as a live system rather than a static database. But the lesson is not tied to one product. The lesson is that the better your context layer, the less generic your AI output becomes.
Your context layer
If you are early-stage, you do not need a perfect CRM. You need a minimum viable context file, a small, maintained source of truth that captures what buyers actually say and do.
Think of it as the marketing memory your company does not yet have.
Capture exact phrases
Start with the words customers use, not the words you want them to use.
- What they say on calls, in DMs, and in replies.
- The exact language they use for pain, urgency, and success.
- Questions they ask before they buy.
Keep these quotes verbatim. Do not clean them up. The rough edges are the value.
Track lost reasons
Closed-lost and “not now” notes are often more useful than wins. They tell you where the market hesitates.
- Too early.
- Missing trust.
- No clear owner.
- Already using a workaround.
- Needs proof of ROI.
These reasons should be tagged consistently. If they are scattered across inboxes, they are not usable context. They are anecdotes.
List common questions
Every prospect asks the same few questions, even if the wording varies. Write them down.
- What does this replace?
- How long does setup take?
- What will break if we switch?
- Why now?
- How do you prove this works?
These questions should shape your pages, demos, and follow-ups. They are your actual buyer curriculum.
How to store it
If you are not ready for a CRM, use a shared spreadsheet, Notion page, or Airtable base. The tool matters less than the discipline.
Create simple fields:
- source,
- date,
- buyer type,
- verbatim quote,
- pain point,
- objection,
- stage,
- next use.
Then review it weekly. A context layer decays if nobody maintains it. The goal is not archival completeness, it is immediate usefulness.
How to use it
Once the context layer exists, feed it into your AI work. The output changes quickly.
Positioning drafts
Ask the model to rewrite your positioning using only customer quotes from the last ten calls. You will get sharper language, and fewer abstract claims.
Objection-aware pages
Build landing pages around the objections you actually hear. If prospects keep saying they already have a workaround, say that directly.
Follow-up emails
Draft follow-up notes in the buyer’s words, not your internal vocabulary. This is one of the highest-leverage uses of AI for first marketers, because it makes outreach feel informed without becoming performative.
Content briefs
Use the context layer to brief articles, webinars, and social posts. The model should not invent the market. It should reflect it.
Why this works
Specificity earns attention. More important, it earns trust. When an AI draft sounds close to the buyer’s actual experience, readers feel understood. When it sounds generic, they move on.
That is the real business case. Better context produces better output, better output produces better resonance, and better resonance reduces the gap between marketing and reality.
For Better Marketing, that is the core standard, useful work grounded in how customers actually buy.
When to upgrade
You should graduate to a real CRM when the manual layer starts slowing you down, or when multiple people need the same source of truth.
Look for these signals:
- you cannot find notes quickly,
- follow-ups depend on one person’s memory,
- leads are getting lost between channels,
- reporting is taking longer than customer work,
- you are spending time retyping context into AI tools.
At that point, choose a CRM that can hold clean fields and support future MCP-style connections. If you later join a 7/28-style demo around connectors and workflows, evaluate it through one lens, how well does this system preserve customer context, and how easily can AI access it?
What to check
When you evaluate a CRM or connector stack, ask practical questions.
- Can we store quotes, objections, and reasons in structured fields?
- Can we tag and retrieve the same themes across accounts?
- Can the system connect to AI tools without custom glue every time?
- Can a non-technical marketer maintain it?
- Will this improve the quality of content, not just the cleanliness of records?
If the answer is no, the system is probably too heavy for your stage.
What people ask
Do I need MCP now?
No. You need usable customer context now. MCP is the connector layer that becomes useful once your systems are worth connecting. If you are pre-CRM, start with structure and consistency first.
Can I do this with ChatGPT?
Yes. Paste in a small, curated context set and ask the model to stay within those bounds. You will get better results than if you rely on a generic brand prompt.
What if our data is messy?
Then clean only the parts you will use this week. A perfect archive is less valuable than a working context file. Start with quotes, objections, and questions, then expand from there.
Takeaway
AI marketing gets less generic when the model can see real customer context. That is the shift behind MCP-connected workflows, and it applies even if you barely have a CRM.
Build a minimum viable context layer now, exact phrases, lost reasons, recurring questions, then use it to brief, write, and follow up. If you do that well, the eventual move to CRM-connected AI will feel like a natural upgrade, not a reinvention.
And more importantly, your marketing will sound like it came from the market, not from a template.