How Do You Get Claude to Target the Right Accounts?

Short answer: Claude cannot pick the right accounts from its training data, because it does not know which companies have the problem you solve this week. To get good targets, give your AI assistant (like Claude or ChatGPT) a ranked list of companies with dated, sourced evidence through MCP, plus your own CRM and call context. Then judge its picks against a short checklist before anyone reaches out.

Why Claude struggles to pick targets on its own

Ask Claude "which companies should I go after?" and you will get a confident, well-organized answer. The problem is what that answer is built on.

• Training data is old. Claude's knowledge stops at a cutoff date. Hiring pushes, launches, leadership changes, and new initiatives after that date are invisible to it.
• It defaults to well-known companies. Without specific evidence, a model leans on the companies it has seen written about most. Those are rarely the best targets for a growing B2B team.
• It cannot see your business. It does not know your closed-won deals, your current pipeline, or which companies your team already worked last month.
• Web search helps, but only partly. Searching one company at a time is slow, and results are uneven. It is hard to search across thousands of companies for the specific evidence that your problem is present now.

The result is a list that looks reasonable and is mostly guesswork. Claude is not the problem. The input is.

What Claude needs to target well

Good targeting depends on two questions: who to work, and why now. Claude can reason about both very well, as long as the answers are in front of it.

1. A ranked list of companies. Companies showing the problem you solve right now, ordered by the strength and recency of the evidence.
2. Dated evidence for each company. The specific change, when it happened, and a source link.
3. Your first-party context. CRM history, open opportunities, past conversations, and call notes, so Claude does not suggest companies you are already working or recently lost.
4. Your definition of the problem. A short description of what a company looks like, publicly, when it needs what you sell.

With that in place, Claude stops guessing and starts reasoning: which of these companies should a seller work first, what is the best way in, and who should get the first message.

How to connect that context through MCP

MCP (Model Context Protocol) is the standard way to give Claude and other AI assistants access to outside tools and data. Instead of pasting spreadsheets into a chat, you connect sources once and Claude can pull what it needs when you ask.

A practical setup looks like this:

1. Connect a source of prioritized companies with dated evidence. This is the who and why now layer.
2. Connect your CRM. So Claude can check ownership, history, and open deals.
3. Connect call and meeting notes, if you have them. So Claude can see what has already been said.
4. Ask specific questions. "Which five companies on my list have the strongest evidence from the last 30 days, and none of them are open opportunities?" works far better than "who should I target?"

For more on what to give Claude, see How Do I Feed Claude Who and Why Now for Outbound? and How to Make Claude Actually Good at Outbound Sales.

A checklist to judge Claude's picks

Even with good context, review the output before acting. Here is a checklist you can use on every company Claude suggests:

Check	What a good pick looks like
Is there a specific reason?	A named change at the company, not a general industry trend
Is it dated?	The evidence has a date, ideally from the last few weeks
Can you open the source?	A link a seller can check in under a minute
Does it tie to your problem?	The reason clearly connects to what you solve
Is it new to your team?	Not already an open deal or recently worked without a new reason
Is the company in your segment?	Size, region, and business model fit who you sell to

If a pick fails the first three checks, treat it as a guess and drop it. If it passes all six, it belongs on the working list.

Prompts that help

A few prompt patterns make Claude more useful once the context is connected:

• "Rank these companies by the strength and recency of the evidence. Show the source and date for each."
• "For each company, explain in one sentence why now, and flag any claim you cannot source."
• "Remove any company that is an open opportunity or was contacted in the last 60 days."
• "For the top five, suggest the best first contact and a first line based on the evidence."

Ask Claude to show its sources every time. If it cannot point to one, the claim should not go into outreach.

Why "just use Claude" is not the whole answer

Claude is very good at reasoning, writing, and planning. It is not a source of fresh, verified company evidence. Teams that treat it as both end up with polished messages to the wrong companies. We wrote more about that in Why "Just Use Claude" Doesn't Work for AI Outbound.

The fix is not a different model. It is giving the model the right inputs. Better context in, better actions out.

Where Syft fits

Syft finds the companies showing the problems you solve right now, with a dated source behind each one, and ranks them so your team knows who to work and why now. It connects to Claude, ChatGPT, and your agents through MCP, alongside your CRM and call data. Your team keeps working in the AI assistant they already use, with the right targets in front of it.

Frequently asked questions

Can Claude find the right companies to target by itself?

It can suggest companies, but without fresh evidence it is mostly working from training data and general knowledge. Those suggestions skew toward well-known companies and miss recent changes.

Does web search fix Claude's targeting?

It helps for researching one company at a time. It does not scale well to finding, across thousands of companies, the ones with dated evidence of your problem this week.

What is MCP and why does it matter for targeting?

MCP is a standard way to connect AI assistants to outside data and tools. It lets Claude pull your prioritized companies, evidence, and CRM context directly, instead of relying on what you paste in.

Does this work with ChatGPT too?

Yes. The same approach applies to any AI assistant (like Claude or ChatGPT) that supports MCP or similar connections.

How many companies should Claude prioritize at once?

Start with the ones your team can act on fully this week. It is better to act on 100% of a prioritized list than to work a long list partway.

How do I stop Claude from making up reasons to reach out?

Give it sourced evidence and ask it to cite the source and date for every claim. Remove any claim it cannot source.

Give Claude the right targets

Syft delivers a ranked list of companies with dated evidence straight into Claude or ChatGPT through MCP. Book a custom demo to see it on your own market.