Your AI tool has found a company that has raised funding, hired a new sales leader and advertised several account executive roles. The email practically writes itself: congratulations on the growth, scaling must be difficult, shall we book a demo?
There is just one problem. Nothing in that research establishes that the company needs what you sell.
This is the awkward gap in AI prospecting: a system can find something true about an account, then wrap an untested assumption around it. The result looks researched. It may still be commercially irrelevant.
The opportunity is to use AI earlier in the decision: not simply to write the message, but to help establish whether there is a worthwhile conversation to start. That means separating what happened, what it might change and what you still need to ask.
TL;DR
- A buying signal is a clue worth investigating, not proof of a problem, budget or active purchase.
- Use AI to assemble evidence and challenge your assumptions before it drafts outreach.
- Require a verified event, a plausible commercial consequence and a relevant question before approving a signal-led message.
- Measure useful conversations and qualified opportunities alongside research cost—not just emails generated.
AI prospecting is moving beyond the writing box
There is a practical reason to revisit this now. HubSpot’s prospecting-agent documentation, updated on 16 September 2026, describes identifying companies through signals, finding relevant contacts and generating outreach using user-defined plays. It also provides a choice between reviewing messages before sending and sending automatically.
That is a documented workflow, not independent proof of better sales results. Nor does the update date establish when each feature first appeared. What matters commercially is the decision it puts in front of a sales team: which parts of prospecting are you prepared to delegate?
If you delegate research, you need to check the evidence. If you delegate account selection, you also need to check the reasoning. Reviewing the final email for tone will not catch every mistake made upstream.
A beautifully written message to the wrong account is still the wrong message.
What is a buying signal—and what does it actually prove?
A buying signal is an observable event or behaviour that may make an account more relevant to investigate. It does not, by itself, establish a buying intention.
For practical B2B prospecting, separate three things that often get bundled together:
- Account fit: the company has characteristics associated with customers you can genuinely help.
- A change signal: something has happened that could alter its priorities, constraints or operating requirements.
- Expressed intent: someone has actually indicated an interest in exploring a relevant solution.
A new office is a change signal. A request to discuss managing access across that office is expressed intent. Treating the first as though it were the second creates false confidence.
Buyer intent data also needs interpretation. Ask what was observed, when, and at what level. An account-level indication of topic research is not evidence that a particular person is running a purchasing project. Keep the claim no stronger than the underlying observation.
This distinction matters for founders choosing their next ten accounts, SDRs working a territory and AEs trying to open a second department. You are deciding where limited attention has the best chance of being useful.
Build an evidence card before you build a sequence
An evidence card is a short account record that separates sourced facts from sales hypotheses. It is an editorially proposed working method, not a vendor score or a validated prediction model.
Keep it compact enough to review without opening a presentation:
- Observed event: what happened, with a source link, publication date and event date where available.
- Account fit: why your product could serve this type of business, including any disqualifiers.
- Possible consequence: the operational or commercial change the event might create.
- Alternative explanation: why that consequence might not exist or might already be handled.
- Relevant role: who is likely to own the process, without assuming they control the budget.
- Next question: what you need to learn before proposing a solution.
The fourth line is where the discipline lives. A company hiring account executives could need onboarding support. It could also be replacing leavers, hiring people into an established programme or using a provider it likes.
Ask AI only for reasons to contact an account and you have designed a justification exercise. Ask it for the strongest reason not to proceed and you have at least created room for a better decision.
Missing evidence should remain missing. Do not turn “no public information found” into “they have no process”. Public visibility and operational reality are different things.
How to use AI to qualify prospecting signals
Start with one narrow commercial question. For example: which suitable software companies are changing their sales organisation in a way that makes onboarding consistency worth discussing?
That is a better research instruction than “find high-intent leads”. It defines the problem you can investigate and gives a reviewer something concrete to challenge.
Collect evidence, not an account biography
Use approved CRM records and relevant public sources: company announcements, current careers pages, product documentation and statements from the business itself. Record when you checked them. A recently indexed page can describe an old event.
Claygent is one relevant research option: Clay documents web-enabled research and structured outputs for account research and qualification tasks. That makes it a candidate for gathering evidence, not a substitute for checking whether its output supports your sales hypothesis.
You can also begin with a company-approved AI assistant and a small set of source extracts. If it cannot access a page, supply the relevant text with its URL and date. Do not ask it to fill the gap from memory.
Use this prompt:
Assess whether this account merits research for our offer: [specific offer and customer fit]. Use only the supplied evidence or sources you can actually access. For each observation, return the source URL, publication date, event date if known and a brief factual summary. Separate verified facts, plausible implications and unknowns. Identify the strongest alternative explanation. Suggest one question for the likely process owner. If the evidence is insufficient, say so. Do not draft outreach yet.
Then open the sources behind the important claims. Check the company identity, whether the event is current, whether a vacancy is still live and whether the description actually says what the summary suggests.
Route the account before drafting
Give the reviewer three possible outcomes:
- Contact: the account fits, the evidence is credible and there is a relevant question worth asking.
- Research or monitor: the account fits, but the timing or commercial connection is unresolved. Record the missing fact and a review date.
- Exclude: a clear disqualifier, contact restriction or ownership conflict makes this outreach inappropriate.
Before contact, check existing opportunities, account ownership, recent conversations and suppression records. Finding a new signal does not erase what another colleague already knows.
A “monitor” decision is not a permanent parking space. Specify what would change your mind. Perhaps it is a confirmed launch date, a role being filled or a previous buyer agreeing that a later conversation would be useful.
From hiring news to a useful conversation: a worked example
The following scenario is hypothetical. It illustrates the method, not a customer result.
You sell sales onboarding software. A suitable SaaS company announces a move into a new market and lists several sales vacancies. An AI-generated draft says its rapid expansion must be creating inconsistent onboarding.
Delete “must”. You do not know that.
The verified facts are expansion and recruitment. The plausible consequence is that new starters may need consistent territory-specific training. The alternative explanation is that the company already has a strong enablement programme. Your relevant contact is the person responsible for onboarding, whose identity you still need to verify.
A more defensible first message would be:
Your announcement mentions the new market launch, and your careers page lists roles supporting it. Is onboarding for those hires staying with individual managers, or moving into a shared programme?
We provide software for structuring sales onboarding. If consistency across teams is under review, would a short comparison of the two approaches be useful?
Only offer that comparison if you can provide something genuinely useful. The message is not improved by replacing one unsupported claim with an imaginary resource.
Notice what the question does. It gives the buyer a straightforward way to confirm, correct or dismiss your hypothesis. It does not ask them to accept a diagnosis from someone who has never spoken to them.
If they reply that the programme is already standardised, update the account record. Do not feed the answer into an objection-handling prompt whose only purpose is to keep the sequence alive. A disconfirmed hypothesis is useful information.
If they say each manager currently handles onboarding differently, you have earned a more specific follow-up: “Where does that variation cause the most difficulty—content, coaching or knowing whether someone is ready?”
That is the transition from account research to discovery. The buyer’s answer, not the AI summary, becomes the basis for the next decision.
The Sales Closer Judgement Test: should this message leave the draft folder?
Use this five-question version of a judgement test before approving AI-supported outreach. It is a practical review checklist, not a scientifically validated assessment.
- Is it true? Can you show the evidence behind every factual claim about the prospect?
- Does it matter? Can you explain how the event could affect something your product genuinely addresses?
- Is it theirs? Is this a sensible conversation for the recipient’s actual responsibilities?
- Why now? Is the timing supported, or are you recycling an old event to manufacture urgency?
- Can they correct us? Does the message invite an answer, rather than present an assumption as a settled fact?
If the first answer is no, remove or verify the claim. If the middle answers are weak, improve the research or choose another account. If the last answer is no, rewrite the message.
This is also a useful coaching exercise. Ask two reps to review the same evidence card independently. Where they disagree, discuss the reasoning rather than asking AI to cast the deciding vote.
The purpose is to make good sales judgement visible enough to teach.
Keep human review where a wrong assumption becomes an external action
For an initial pilot, let AI gather and organise evidence; have a person approve the contact decision and message. In HubSpot, the documented “Review before sending” option provides a relevant control. Confirm availability, permissions and credit requirements for your account before building around it.
Set operational boundaries outside the prompt wherever your systems allow. Restrict which accounts can enter the workflow, who can approve messages and which records must be excluded. A sentence telling AI to be careful is not the same as a sending restriction.
Use company-approved systems and the minimum information necessary for the task. Keep irrelevant personal details out of research and messaging. Your existing privacy, contact-permission and suppression processes still need to apply; a signal-led workflow should not bypass them.
When an error appears, correct the account evidence as well as the draft. Otherwise the next message may reproduce the same false premise in different words.
How to measure whether AI prospecting improves lead quality
Run a small, bounded pilot in one segment. Choose an account set your team can review properly, and compare it with similar accounts worked through your existing approach. Keep the offer, qualification criteria and observation period consistent. Do not select only the most promising accounts for the AI group and then credit the tool for the difference.
Track the work from research to outcome:
- Research quality: how often a reviewed evidence card contains an unsupported or materially outdated claim.
- Efficiency: research, review and correction time per account, plus tool cost.
- Conversation quality: the proportion of contacted accounts producing a reply that confirms a relevant issue or starts a substantive discussion.
- Commercial progression: held meetings and opportunities accepted against your normal qualification criteria.
- Negative outcomes: complaints, opt-outs, duplicate contact and buyer corrections of your assumptions.
Define those measures before the pilot. A polite “thanks” is not a useful conversation. An accepted calendar invitation is not a held meeting. A meeting is not automatically an opportunity.
Use accounts contacted as a consistent denominator where appropriate, and record multiple contacts within each account separately. Otherwise a heavier contact cadence can make a comparison look stronger without showing that account selection improved.
TSC’s guide to prospecting metrics for SDRs provides broader measurement context. Here, the additional question is whether the evidence-led decision earned its cost.
Small samples are directional, not proof. If the workflow saves research time but generates more corrections from buyers, investigate before expanding it. If it produces fewer messages but more qualified conversations at an acceptable cost, reduced activity may be a sensible trade.
What to change before your next prospecting session
Do not begin by rewriting every sequence. Take five accounts already in your queue and build an evidence card for each. Look for the point where a fact became an assumption.
- Replace vague “high intent” labels with the observation and its date.
- Ask AI for an alternative explanation before asking it for an email.
- Give uncertain accounts a research decision, not an automatic sequence.
- Use the buyer’s response to update your understanding, not merely your next line.
You do not need perfect knowledge before contacting someone. You need enough evidence to ask a useful question, and enough honesty to admit what you do not yet know.
To put that into a wider prospecting routine, get the free SaaS Prospecting Scorecard and join The Sales Closer’s subscriber community. Use it to identify what needs attention across your system, then bring this question into the conversation: which signal looked promising until you checked what it actually meant?
AI can help you find the clue. The quality of the next question is still your responsibility.