Research finds the client. A decision layer decides if that client is even worth a message.

Marketing has changed more for me in the last few months than it did in the years before that.

Grok Bot is a big reason. It let me stop treating research and outreach as two separate jobs I had to babysit. I could point an agent at a perfect client profile, let it pull the data through Cursor, turn that data into an email, and send the outreach.

We were doing that roughly 100 times per day.

That already felt like a different business. Then Jev showed up, and the system changed again.

Jev is not another research engine. Jev is the in between.

Cursor still does the research. Jev tells us what to move forward on and what to leave alone. Only after that decision do we send the outreach email.

That one extra seat in the middle is the biggest marketing use case I have found with Jev so far.

The first shift: research and outreach at volume

Before this stack, the bottleneck was obvious.

  • Someone had to find the company.
  • Someone had to decide if they looked like a fit.
  • Someone had to write the email.
  • Someone had to send it.

Each of those steps cost time. Most teams skip the second one when they get busy. They research a list, write a template, and fire. Volume goes up. Fit goes down. The inbox fills with noise and the close rate quietly dies.

Grok Bot broke the time problem.

I started having Grok Bot do extensive research through CLI into Cursor. The job was simple to describe and expensive to do by hand:

  • Find my perfect client.
  • Pull the data that proves they might be a fit.
  • Turn that data into an email.
  • Send the outreach.

Roughly 100 times a day.

That is not a content trick. That is an operating system for outbound. The agent can keep working while I am in the office, with family, or asleep. The research does not wait for me to open a tab.

If you only take one thing from the first chapter of this, take this:

The win was not AI wrote an email. The win was a loop that could run at a volume a human team cannot sustain without getting sloppy.

The problem volume creates

Once you can research 100 prospects a day, a new problem shows up.

You can send 100 emails a day. That does not mean you should.

A lot of those companies will look fine on the surface. Website exists. They are in the right industry. They are in the right city. They have a phone number. That is enough for a bad list. It is not enough for a good one.

If every researched lead becomes an email, you are paying for the research twice. First in compute. Second in reputation. People remember a message that had no reason to land in their inbox.

The old fix was a human filter. A person skimmed the research and said yes or no. That works until the pile is 100 deep and the person is also running a company.

The new fix is a decision layer with a written standard.

That is where Jev sits in my stack.

The second shift: a gate before the send

Jev is the in between.

  • Cursor does the research.
  • Jev tells us what to move forward on and what not to move forward on.
  • Once we have that answer, we send the outreach emails.

We built a criteria set of roughly 60 questions. Those questions qualify a potential client immediately. The output is not a long essay. The output is a decision. Reach out, or do not.

That is the whole point.

Research can be long. The gate should be short. If the criteria cannot produce a clear yes or no, the criteria is not finished.

I am not going to invent a magic close rate here. I will tell you what the system actually does.

  • It stops the send when the prospect fails the standard.
  • It lets the send through when the prospect clears the standard.
  • It does that before anyone spends another minute rewriting the email.

How the loop works in plain language

Think of three seats.

Seat one is research.

This is Cursor, driven by Grok Bot through CLI. It pulls public data on the company, the offer, the market, the site, the content, the obvious signals that this might be the kind of client we want.

Seat two is judgment.

This is Jev. It does not go hunt a new list. It looks at the research against a fixed set of questions and returns a go or a no go.

Seat three is action.

This is the outreach email. It only exists when seat two said go.

If you mix those seats, the system gets sloppy. If research also decides, you get optimistic emails. If the email writer also decides, you get pretty messages to bad fits. If a human has to reread every file, you are back to the old bottleneck.

Keep the seats separate.

What a 60 question gate is actually doing

Sixty sounds like a lot. It is not sixty essays. It is sixty checks.

A good check is binary or close to it.

  • Is this company in the market we serve?
  • Do they already buy the kind of work we sell?
  • Is there a public sign they are active right now?
  • Does the size match what we can deliver?
  • Is the contact path real?
  • Does anything in the research say this would be a waste of their time and ours?

You group those checks so the model is not guessing from vibes.

  • Fit. Who they are.
  • Timing. Whether now is a sane moment.
  • Capacity. Whether the work is the size you actually take.
  • Proof. Whether the research found something real instead of a guess.
  • Risk. Whether outreach would look reckless or off brand.
  • Path. Whether you can even reach a human without playing games.

When those buckets are written down, Jev is not being creative. Jev is scoring the file against your rules.

That is why this belongs in marketing. Marketing dies when the standard lives in one person's head and disappears the minute that person gets busy.

How to use this in your own marketing

You do not need my exact stack to steal the shape.

You need three things.

  1. A research step that can pull a real file on a company.
  2. A written gate with questions you would actually use if you were reviewing the file yourself.
  3. An action step that only runs after a yes.

Start smaller than 60 if 60 would make you stall. Ten sharp questions beat sixty vague ones. Grow the list when you see the same bad lead slip through twice.

Write the questions in your voice. If you would not reject a lead for that reason in real life, do not put it in the gate.

Then pick one action. One. For me that action is the outreach email. For you it might be a LinkedIn note, a call list, a proposal draft, or a content brief. The action is allowed to be ambitious. The trigger is not. No yes, no action.

Example: local service marketing

Say you run ads and sites for roofers, fence companies, pools, or shops in one city.

Research can find every contractor with a site and a Google profile. That list is huge and mostly useless.

The gate should ask things like:

  • Are they in the service area you actually sell into?
  • Do they already spend on ads or content, or is the presence dead?
  • Is the site so broken that they feel the pain, or so new that they just paid someone else?
  • Do they look like a price shopper only, or like an owner who cares about leads?
  • Is there a real person to contact, not a form that goes to nowhere?

Yes means you write a specific note about what you saw. No means you do not burn the domain on a company that was never going to buy.

Same research muscle. Better send list.

Example: agency retainers

Agencies love volume. Volume is how retainers turn into cheap project work.

Research can build a beautiful deck on any brand. That is not a reason to pitch them.

The gate should protect the retainer.

  • Do they already have a marketing person, or are they the owner doing it at midnight?
  • Is the work the kind you keep, not the kind you resent?
  • Is there a budget signal in public, or only a pretty logo?
  • Have they churned agencies in public ways that tell you the relationship will be ugly?
  • Can you name the outcome you would be hired to move?

If the file cannot clear those questions, the pitch is a vanity project. Jev's job is to say no before the deck gets built.

Example: product and ads

If you sell a product, the same pattern works on campaigns, not just people.

Research pulls the live ads, landing pages, offers, and comments. The gate asks whether this campaign looks like your buyer, your price, your promise, and your next test. Action is only the next ad, the next email, or the next page.

Most teams generate more ad variants than they can judge. That is the same disease as 100 outbound emails with no filter. Creation got cheap. Judgment did not. Put judgment in a seat that only does judgment.

Example: content as marketing

Content teams generate ideas all day now. That is not a strategy.

Research can gather what is working in your space this week. The gate asks whether the idea serves the buyer you sell to, whether you have proof, whether it can be made this week, and whether it would embarrass you if it were the only thing a client saw. Action is the outline, the shoot list, or the post. Not another idea pile.

The point is the same in every example.

  • Do not let the generator decide.
  • Do not let the sender decide.
  • Give the decision its own step.

How to write questions that actually qualify

A weak question is: Is this a good lead? A strong question is one a junior hire could answer from the research file with almost no debate.

Use questions that force a source.

  • What page proves they sell the thing we help with?
  • What public activity is dated recently enough to matter?
  • What would make this a bad client even if they said yes?
  • What would we offer in the first email that is specific to them?
  • If we cannot name that offer, why are we emailing?

Kill questions that reward fluff. If the model can answer yes by being polite, the question is broken.

Add a hard stop question. If this one fails, the rest do not matter. Geography is a common hard stop. So is industry. So is: we do not have a way to reach a human.

Review the no pile once a week at first. If a no looks like a company you would have wanted, the question set is wrong. Fix the questions. Do not start overriding the gate by mood.

What this is not

  • This is not a claim that AI should send everything while you sleep with no standard.
  • This is not a claim that research equals a qualified lead.
  • This is not a tutorial that invents features Jev does not have in my workflow.

In my workflow the jobs are clear.

  1. Grok Bot plus Cursor pull the file on the perfect client.
  2. Jev runs the criteria and says move forward or do not.
  3. Outreach happens after that.

That is the system. The rest is discipline. Keep the standard written. Keep the send behind the standard. Keep raising the questions when bad fits sneak through.

Why this matters right now

Marketing is changing at an exponential rate.

A few months ago the leap was an agent that could research and draft at a pace a person cannot match. That already changed how I work. Grok Bot made that real.

A month later the leap is a layer that can say no.

That sounds smaller. It is not. The expensive part of marketing is not finding names. The expensive part is talking to the wrong ones, training your team on sloppy leads, and teaching the market that your messages are generic.

When creation is cheap, taste and standards become the product.

Jev is how I put a standard in the middle of a machine that can otherwise run all day.

If you already have research and outreach, add the missing seat.

  • Write the questions.
  • Make the send wait.

That is the whole upgrade.