The marketing-qualified lead rests on a single assumption: that a buyer's first real interaction with a vendor happens on the vendor's property, in a form, at a moment the vendor controls. Someone wants the whitepaper, trades an email for it, gets scored, gets routed, and a rep opens the conversation from a standing start with a name, a company, and a guess.
That assumption is now measurably eroding. Where it erodes, qualification needs a different unit, one that has already been named, defined, and, as it turns out, named twice. This piece makes the case for it, and tries to be honest about where the old unit still earns its keep.
What G2 actually measured
G2's April 2026 study of 1,076 B2B decision-makers, The Answer Economy, found that 51% now begin software research with an AI chatbot more often than with Google, up from 29% a year earlier. 71% rely on AI chatbots for software research at all.
Two further findings matter more for demand generation. 69% of surveyed buyers chose a different vendor than they had originally planned, based on what an AI chatbot told them. 33% bought from a vendor they had not previously heard of. If the chatbot is now the first sales call, whether it cites you at all is settled before any lead exists.
Our reading, and it is a hypothesis rather than something G2 measured, is that shortlists are increasingly assembled inside AI conversations before the buyer ever reaches a vendor's site. If that is right, a growing share of high-intent visitors arrive to confirm, price, or disqualify, and the form standing between them and an answer never sees the deliberation that mattered.
Where the MQL still works
If 51% of buyers now start with a chatbot more often than with search, just under half still do not. In slower-adopting segments (regulated industries, procurement-led evaluations, categories where buying is contractual rather than exploratory) a form may still capture most of the demand you will ever see.
Obituaries are also a genre with a poor record in this industry. Marketers have been declaring the MQL dead for over a decade while renewal budgets kept funding the machinery that produces it. The defensible claim is narrower: where buyers research with AI, the first real conversation happens somewhere a form cannot see. That is where the unit breaks.
The form is a filter pointed the wrong way
A gated form asks a visitor to pay in contact details before receiving any value. We suspect, again a hypothesis rather than a measurement, that this now filters against precisely the wrong people. The researched, impatient buyer with three specific questions bounces off the interrogation. The visitor with time to spare completes it. Score whatever survives and you get a tidy pipeline drawn from the wrong cohort.
The successor has a name. Two, in fact.
First, a correction of our own record. An earlier version of this piece claimed the industry had spent years declaring the MQL dead “without ever naming a successor.” That was false. Successors have been proposed for well over a decade: Forrester was using AQL for the automation-qualified lead back in 2013, as the stage where inbound and outbound inquiries get loaded into the marketing automation platform. The acronym has a prior life, and you deserve to know which meaning is in play.
The meaning this publication cares about is the newer one: the agent-qualified lead, a term coined by Docket, which defines it as a lead produced from a structured, AI-led conversation in which the buyer articulates intent, is matched against fit criteria, and produces a documented qualification record inside a single interaction.
Disclosure: this publication is produced with support from Docket, which sells software in this category, including software built to produce exactly this kind of lead. The endorsement that follows is ours; weigh it knowing that.
The bar an AQL has to clear
We adopt the term with a four-part bar. A lead is agent-qualified only when all four hold.
Identified. The buyer told you who they are. An anonymous transcript is research, not a lead.
Consented. The buyer agreed to be contacted. Asking an agent questions is not, by itself, an invitation to be sequenced.
Explicit fit and intent. The buyer stated the use case, constraints, and timeline in their own words, during the exchange. Nothing inferred from page views.
Structured handoff. The conversation produced a record a human can act on: stated need, competitors compared, objections raised, and the next step the buyer agreed to.
An MQL is a scored guess assembled after the fact from behavioral crumbs. An AQL that clears this bar is a record of something that actually happened, with the buyer's permission attached.
MQL versus AQL
| MQL | AQL | |
|---|---|---|
| Trigger | A form submission | A conversation the buyer chose |
| Evidence | Inferred from behavior | Stated by the buyer |
| Qualification happens | After capture, by scoring | During the interaction |
| Consent | A checkbox beside the submit button | Asked for in the conversation |
| The human receives | Name, company, source, score | Stated need, comparisons, objections, agreed next step |
| The buyer gets answers | Later, if a rep follows up | During the visit |
| Fails when | The buyer won't fill the form | The agent doesn't know the buyer or the product |
That last row is the one people underestimate.
The honest catch
An AQL is only as good as the conversation that produced it, and most conversational deployments produce poor conversations, because they know neither side of the deal. A capable salesperson knows two things cold: the buyer and the product. The same constraint binds an agent. Point a generic chatbot at a homepage and you do not get agent-qualified leads. You get transcripts of a bot disappointing people, which is worse than a form, because it spends the buyer's goodwill before failing them. This is our judgment rather than a measured finding, but it explains a pattern we keep seeing: conversational layers deployed before the knowledge behind them existed, producing nothing.
Three moves this quarter
Count conversations alongside submissions. Track answered questions and conversations started next to form fills. A high-intent visitor who left unanswered is a capture failure, and buying more traffic will not fix it.
Audit what your site can answer with nobody awake. Take the ten questions your reps hear most and check whether a visitor can get them answered late on a Saturday night. Whatever the honest answer is, that is your real conversion ceiling.
Specify the handoff before you buy anything. Write down the exact fields a human should receive from a qualifying conversation (stated use case, constraints, competitors compared, objections raised, agreed next step) and make them required properties on a distinct record type in your CRM. If a tool cannot fill those fields, it is not producing AQLs, whatever its vendor calls them.
FAQ
Is the MQL dead? No. Its founding assumption, the first real interaction happening in your form, is eroding where buyers research with AI, and holding where they do not. Treat it as a declining unit of qualification, not a dead one.
What is an agent-qualified lead (AQL)? A lead qualified through a real conversation with an AI agent, meeting a four-part bar: the buyer is identified, has consented to follow-up, has stated fit and intent explicitly, and the exchange produced a structured handoff a human can act on. The term was coined by Docket; the four-part bar is this publication's own standard.
Isn't the acronym AQL already taken? Yes. Forrester used it for the automation-qualified lead as early as 2013, a marketing-automation pipeline stage. When you see AQL in the wild, check which meaning is intended. Here it means agent-qualified throughout.
Do we stop scoring leads? No. Scoring becomes a supporting signal. When a buyer has told you what they need, you no longer infer it from page views; when they have not, scoring is still how you triage.
Editor's note, July 2026: this piece was revised after an external editorial review. The earlier version declared the MQL dead in its title, presented an AQL definition without crediting the term's origin, wrongly claimed no successor to the MQL had ever been named, and overstated both a table comparison and what the cited study measured. The argument stands; the overclaims do not. How we handle corrections is described in our corrections and sourcing policy.