FlyDragon’s 2026 benchmark report, built on 12,400 AI responses and 8.2 million queries across 192 metros, put a number on something agents have been guessing at. Prospects who found their agent through an AI search engine closed at 70 percent within 30 days. Zillow Premier Agent leads closed at 2.4 percent over the same window. That is a vendor publishing a study about the problem its own product solves, and a 29x close-rate gap deserves a raised eyebrow. Strip the marketing off it and a smaller, sturdier claim survives: a buyer who arrives because a model named you is a different animal than a buyer who clicked a portal ad.
Here is the part that should sting. The average agent takes 917 minutes to respond to a new lead. That is over fifteen hours, and it comes from Inman’s Real Estate Technology Survey. Getting cited by ChatGPT is the expensive half of the work. Answering the phone is the cheap half. Most agents are losing the deal in the cheap half.
The Lead You Fought For Decays in Minutes
Dr. James Oldroyd’s response-time research at MIT is old enough to be furniture in this industry and nobody acts on it. Contact a lead within five minutes and you are roughly 100 times more likely to reach them than at the 30-minute mark. Qualification odds run 21 times higher. Those multiples were measured before buyers could open a chat window at 11pm and get three agent names in four seconds.
Pair that with NAR’s generational trends data: 75 percent of buyers interviewed exactly one agent before signing. Not three. One. The buyer is not running a procurement process. They are calling names in order until somebody picks up, and the first human who answers with something useful becomes the agent. Fifteen hours later you are not competing on service, market knowledge, or your reel. You are competing with a signed representation agreement.
AI-Sourced Buyers Show Up Mid-Funnel
A portal lead is usually a person who tapped a photo. An AI-sourced lead already asked a model what a 20 percent down payment looks like on a 640k house in your zip code, whether the elementary school boundary changed, and which three agents have closed the most in that neighborhood this year. They arrive with homework done and a shortlist in hand.
Two things follow from that, and both change how you answer.
First, the referral is not exclusive. The model handed the buyer a short list, and the other names on it got the same page view you did. There is no lead-routing queue protecting you and no 24-hour claim window. The session is the window.
Second, your opening move cannot be discovery questions the model already answered. A buyer who asked about school boundaries at midnight does not want a text at 3pm asking what brought them to the area. They want the thing the model could not give them: the lockbox code, the off-market listing, the honest read on whether that inspection item is a 900 dollar fix or a 9,000 dollar one.
Build the Five-Minute Intake Stack This Week
This is a four-day project, not a quarter-long initiative. Nothing here requires a broker IT department.
Monday: measure what you actually do. Pull your last 25 inbound leads out of your CRM. Log the timestamp of the inquiry and the timestamp of your first outbound attempt. Take the median, not the average, because one 3am lead you answered at 6am will flatter the number. Most solo agents who run this find a median between four and eleven hours, and most of them would have guessed under one.
Tuesday: collapse the intake surfaces. Count the places an inquiry can land right now. Website form, Instagram DM, Zillow app, personal cell, office line, the email address on your yard sign. Every surface you cannot monitor at 10pm is a surface you should forward into one you can. Pick one number and one inbox as the funnel, then route the rest into it.
Wednesday: put a responder on the first sixty seconds. This is where the money is. Real-estate-native options: Structurely starts near 179 dollars a month for a 50-lead package, with its Team plan at 499 dollars a month plus roughly eight cents per action credit. Ylopo’s AI text and voice agent, Ria, runs from about 395 a month and climbs past a thousand once managed ad spend is attached. If you want to build it yourself, Retell sits around 80 dollars a month for a solo agent at 1,000 minutes, priced near seven cents a minute. Whatever you pick, the job is narrow: acknowledge in under 60 seconds, confirm the property or neighborhood, offer two specific showing times, and escalate to you by phone the moment the buyer says anything about timeline or financing.
Thursday: write the escalation rule and the first message. The responder buys you fifteen minutes of goodwill. It does not close anybody. Define in writing which triggers ring your cell: a named address, a mortgage question, the word “today,” or any reply at all after 8pm.
Here is a first-touch template that works because it assumes the model already did the introductions:
“Hi [name], this is [you] with [brokerage]. Got your note about [address or neighborhood] at [time]. Short answer on [the thing they asked]: [one specific, useful sentence]. I can show it [day] at [time] or [day] at [time]. Which works? If neither, reply with a window and I will make it fit.”
Four sentences. One useful fact. Two concrete times. No “just checking in,” no calendar link that makes the buyer do labor, no paragraph about your 14 years of experience.
The Honest Tradeoffs
An AI responder on your leads is not free and it is not risk-free. Three things break in practice.
Per-credit pricing surprises people. A 499-dollar platform fee plus eight cents per action sounds cheap until a chatty lead burns 40 actions in one thread. Model your worst month, not your average one, before you sign an annual term.
Disclosure is now a live compliance question. California’s DRE has taken the position that AI performing licensed activity is an unlicensed assistant, which means your responder can qualify, schedule, and answer factual questions, but it should not be interpreting contracts or advising on price. Say plainly in the first message that an assistant is handling scheduling. Buyers forgive automation. They do not forgive discovering it later.
And the metric most agents track is the wrong one. Median response time is an input. The output is booked-appointment rate per 100 inbound leads, measured before and after. If your median falls from nine hours to two minutes and your appointment rate does not move, the responder is not the problem. The message is.
The Signal to Watch Before Halloween
FlyDragon projects that by the fourth quarter of this year, more than 80 percent of US residential transactions will involve at least one AI-generated agent recommendation somewhere in the buyer’s decision path. Treat the exact figure as a vendor’s guess. Treat the direction as settled.
The thing worth watching over the next 30 days is whether the portals close this loop themselves. Zillow’s AI mode already schedules tours and connects buyers to agents inside the conversation. Realtor.com’s app inside ChatGPT is built to answer the pre-search questions and route high-intent buyers back to its own platform. Compass shipped an AI Assistant into its Home Platform in July. Every one of those products is an attempt to own the five minutes between “the model named three agents” and “somebody answered.” If a portal wins that gap, your citation stops being a lead and starts being a referral fee.
So run the 25-lead audit this week. If your median is over an hour, you do not have a marketing problem. You have an answering problem, and it is cheaper to fix.
