What is a NO_HOOK rule in AI SDR research?

A NO_HOOK rule is one line in your research prompt that lets the AI say "I found nothing" instead of inventing a reason to email someone. When the research step can't find a specific, recent, checkable detail about an account, it returns NO_HOOK and the record goes to a person or a different queue. No hook, no personalized opener.

Where does the NO_HOOK rule come from?

It comes from Clay's 2026 guide to AI SDRs. The guide's account-research prompt asks for one specific, recent, verifiable detail about a company (a launch, a new market, a hiring spike in one function, a leadership change). Then it ends with this line: if nothing specific is found, return "NO_HOOK" so a human reviews it.

Clay says that line matters more than the rest of the prompt, and that the prompt "must be allowed to say it found nothing." I agree. Without that permission, a model asked for a hook will hand you one.

Why do AI SDRs invent hooks?

Because the prompt rewards a filled field and never rewards an honest blank.

Ask a model for "one relevant detail about this company" and you'll get one every time. Sometimes it's real. Sometimes it's a stretched reading of an old press release. Sometimes it's made up. The buyer can check in ten seconds, and a fake hook tells them you didn't look.

Clay's guide names the damage: an agent that occasionally hallucinates a fact about an account burns trust with reps and prospects. Reps stop believing the list. Prospects learn to ignore the sender.

What does the NO_HOOK rate tell you?

It tells you how much of your list has a real reason to hear from you this week. That's a Signals number most teams never measure.

Ehrenberg-Bass Institute research for the LinkedIn B2B Institute finds that about 95% of B2B buyers are out of market at any given time. So expect a lot of NO_HOOK on a cold list. If most of a list comes back empty, the research step is telling you the truth about the list. Inventing hooks for those records hides the problem. It doesn't fix it.

Track the rate by list source:

List source	What a high NO_HOOK rate usually means
Static firmographic pull	Fit with no timing. Add a signal filter before research.
De-anon visitors (RB2B, Vector)	Visits without context. Check which pages and how recent.
G2 category or competitor traffic	Should run low. If it's high, the prompt is too strict or the data is stale.
Job posts or trigger lists	Should run low. If it's high, the trigger is too broad.

A falling NO_HOOK rate means your list is getting better. A rate near zero on a cold list usually means the model is making things up.

How do I set up a NO_HOOK rule?

Five steps.

1. Define what counts. Specific, recent, and checkable. Name the types you accept: a launch, a new market, a hiring spike in one function, a leadership change, a public statement of a priority. Put a date limit on "recent."
2. Make NO_HOOK an allowed answer. Write it into the prompt as a valid output. "If you can't find one, return NO_HOOK." Models follow the options you give them.
3. Require the source. The research step returns the URL with the hook. No URL, treat it as NO_HOOK.
4. Route the blanks. NO_HOOK records skip the personalized opener. They go to a person, to nurture, or off the list.
5. Spot-check the hooks. Every week, open 20 hooks and click the source. If any are wrong, tighten the prompt before you scale it.

What do I do with NO_HOOK records?

Three options, in order.

• Drop them for this cycle. They aren't in a window. Run them again next month.
• Send them to nurture. Out-of-market nurture keeps you known without an ask, so you're on the list when the window opens.
• Hand-research the best-fit ones. A person spends five minutes and often finds what a prompt missed, like a podcast interview.

What you don't do is send a generic "personalized" note anyway. A note that could have gone to anyone does not go. The bar is a first-touch note that could not go to anyone else, and NO_HOOK is how you hold that bar before the draft even exists.

Does a NO_HOOK rule slow outbound down?

It cuts sends, and more of what's left is worth reading. Clay's guide argues the real lever is how many researched accounts a rep can work in a week. A NO_HOOK rule keeps "researched" honest. Your domain also carries less dead weight, and domain burn is what noise costs you.

What should I do this week?

1. Open your research prompt. If it can't return "nothing found," add the NO_HOOK line.
2. Require a source URL for every hook.
3. Run 50 accounts and count the NO_HOOK rate.
4. Click 20 hooks. Count how many were real and recent.
5. Route NO_HOOK records to nurture or review, never straight into the sequence.

If your AI never says "I found nothing," some of its hooks are made up. Give it permission to come back empty, then count how often it does.

Start Signals, Convert, Grow

FAQ

What is NO_HOOK in Clay?

It's an output in the account-research prompt from Clay's guide to AI SDRs. When the research step can't find a specific, recent, verifiable detail, it returns NO_HOOK so a person reviews the record instead of the model inventing one.

Is a high NO_HOOK rate bad?

Not on a cold list. Most buyers aren't in market, so a high rate is honest. A high rate on a signal-based list, like G2 category traffic or trigger events, means the signal or the prompt needs work.

Can I use a NO_HOOK rule outside Clay?

Yes. Any research prompt can return a fixed token when it finds nothing. Add a route for those records and you have the rule.

Should AI write an opener when there's no hook?

Not a personalized one. Clay's guide treats first-touch copy as draft-then-approve with a person on the send. With no hook, there's nothing true to personalize, so the record goes to nurture or review.