AI-Mediated B2B Buying: How to Run Outbound When the Shortlist Forms Without You

The shortlist forms before you know the deal exists

Two pieces of research published this summer, read together, should change how you plan Q4 outbound.

The first: IDC research published in August 2026 found that eight in ten B2B technology buyers are already using AI agents as part of their purchasing process. Not planning to. Not piloting. Using.

The second: G2's 2026 Buyer Behavior Report: The Evaluation Maze, based on a survey of more than 1,000 B2B software buyers plus interviews with 50+ sales and marketing leaders, found that 82% of buyers have sourced software recommendations from an AI chatbot in the last 24 months — and that evaluation has now overtaken research as the longest stage of the buying journey for the first time.

AI-mediated B2B buying is no longer a 2028 forecast you can put in a board deck and ignore. It is the current state of your pipeline, and it has a very specific, very inconvenient consequence for outbound teams: the part of the buying journey that used to leave footprints on your website now happens inside a chatbot that tells you nothing.

If your prospecting motion depends on knowing who is in-market — website visitors, content downloads, G2 category page views, third-party intent spikes — a meaningful share of that signal is quietly draining away. Buyers are not doing less research. They are doing it somewhere you cannot instrument.

This post covers what the new data says, why it breaks the intent-data playbook most teams bought between 2022 and 2024, and which signals still reliably tell you someone is worth reaching out to.

What the new research actually says

It is worth being precise here, because a lot of the commentary on this topic rounds real numbers into vibes.

Three structural shifts fall out of that table.

Discovery compressed. Evaluation expanded.

G2's framing is that AI solved one constraint and created another. Finding vendors used to take hours of browsing, analyst reports, and peer asks. Now it takes a prompt. But the friction did not disappear — it moved downstream into evaluation, where buyers compare finalists, validate proof, scrutinise pricing, survive a security review, and try to get a signature.

For sellers, that means the window where you could earn a place on the shortlist through outreach has shrunk, while the window where you have to defend your place has grown.

Agents research. Humans still decide.

This is the nuance most people miss. G2 found that only 47% of buyers would let an agent conduct research and make recommendations with humans retaining all final authority, just 9% are comfortable letting an agent execute purchases inside approved guardrails, and a rounding-error 2% would allow purchases without pre-approval.

The most common agent use cases sit squarely in evaluation: understanding total cost of ownership (51%), building shortlists (51%), researching solutions (49%), and evaluating shortlisted vendors (46%).

So agents are analysts, not buyers. You are not selling to a bot. You are selling to a human who arrives pre-briefed by one — often with a frontrunner already in mind and a set of objections already formed.

Finance became a second buying committee.

Finance involvement jumping from 31% to 46% in a single year is the kind of move that quietly wrecks forecast accuracy. Nearly half of buyers report a CFO reversing an already-approved deal. Among organisations with dedicated token or LLM budgets, that climbs to 54%.

And buyers who have been through a late-stage veto behave differently afterwards: three in four expect positive ROI within six months of signing, and they push for sub-12-month contracts at more than double the rate of everyone else (40% vs 18%).

Why this breaks the intent-data playbook

Most modern outbound stacks were designed around a set of assumptions that AI-mediated buying quietly invalidates.

Assumption one: research leaves a trail on your property. The 2022-era playbook said buyers would visit your pricing page, read three blog posts, and download a comparison guide before you ever heard from them. De-anonymise that traffic, score it, route it. But when a buyer asks a chatbot to compare six vendors in a category and returns a scored table, they may never touch your site during the phase where the shortlist is being set. Anonymous research is now genuinely anonymous.

Assumption two: third-party intent equals in-market. Category-level intent data works by observing content consumption across a publisher network. That network was built for a world where humans clicked through to articles. As more of that consumption is summarised by an assistant, the observable surface thins. Intent data is not dead — G2's own finding that review sites are now the top shortlist-shaping source (38%, narrowly ahead of chatbots at 37%) shows first-party review-site behaviour is still enormously valuable. But breadth is narrowing, and lag is growing.

Assumption three: a form fill marks the start of the journey. Increasingly, a demo request marks the middle. By the time a request lands, the buyer has often already built a shortlist, formed a hypothesis, and started assembling an internal business case. Treating that lead as top-of-funnel — a slow nurture sequence, a discovery call that re-covers ground they already researched — actively loses deals to whoever engages like the deal is already live.

Assumption four: volume compensates for weak targeting. It does not, and the arithmetic is getting worse. Wider outbound volume against a buyer population that increasingly filters, delegates, and self-serves is how teams end up with more sends and less pipeline. That is a maths problem no amount of sequencing software solves.

The uncomfortable conclusion: the funnel is not shrinking, it is going dark earlier. The dark funnel that marketers have complained about for a decade just got substantially darker.

The signals that survive AI-mediated buying

Here is the good news, and it is the whole reason signal-based outbound is having a moment.

AI agents summarise published information. They do not — and cannot — summarise the messy, public, human behaviour that surrounds a buying decision. That behaviour is still observable, still timestamped, and still a far better predictor of intent than a category-level score.

The signals that keep working are the ones tied to a specific person doing a specific thing at a specific moment:

• Someone publicly asking for tool recommendations. A post on LinkedIn, Reddit, or X saying "what are people using for X?" is the purest in-market signal that exists. It is also the one most likely to be answered by a chatbot if you are not there first.
• Someone complaining about an incumbent. Frustration posts, migration threads, "we're finally ripping out [tool]" updates. Switching intent, publicly declared.
• Hiring signals. A company posting for a role — an RevOps hire, a first SDR, a compliance lead — is telling you what they are about to build, and therefore what they are about to buy. Job posts are one of the few forward-looking signals available for free.
• Funding and expansion events. A round, a new office, a new market. Budget arrives and buying committees form. The window is weeks, not quarters.
• Engagement on relevant content. People who liked, commented on, or shared a post about the exact problem you solve — including posts by your competitors — have self-identified as caring about the topic today.
• Profile views. Someone in your ICP looking at your profile after you posted something is a warm signal you own outright, and one no AI intermediary sits between.
• Job changes. New leaders re-evaluate the stack in their first 90 days. It is the most reliable recurring buying trigger in B2B, and it is fully public.
• Community activity. Slack groups, subreddits, niche forums where practitioners ask real questions before they ever open a vendor site.

Notice what these have in common. They are all human, public, and time-stamped. They are not intent scores inferred from aggregate content consumption. They are individual observable acts, which is exactly what remains legible when research moves inside an assistant.

Old intent model versus surviving signal model

This is the thesis behind Updately and behind signal-based selling generally: stop guessing which accounts are in-market and start watching what people actually do. Capture profile views, post engagers, competitor mentions, hiring signals, and pain-point posts across LinkedIn, Reddit, and X; enrich and score them against ICP; and reach out while the signal is still warm rather than three weeks later when it has aged into noise.

What to do about it this week

Strategy is cheap. Here is the operational version.

1. Re-baseline your outbound assumption from "cold" to "already shortlisted"

Rewrite your opening messages on the assumption the prospect has already asked an AI assistant about your category and has a rough map of the landscape. That kills two habits immediately: explaining what category you are in, and asking questions they have already answered for themselves.

What replaces them is a specific, observable reason you are writing. "I saw your post about X" beats "I wanted to introduce our platform" by an order of magnitude, and the gap is widening because the second message is now competing with a chatbot that explains categories better than you do.

2. Instrument the signals you can still see

Audit where your pipeline signals currently come from. If more than half your prospecting triggers are website-based or third-party-intent-based, you are exposed to the erosion described above. Add at least three human, public signal sources this quarter — post engagement, hiring posts, and community pain-point posts are the fastest to stand up.

Track a simple metric: time from signal to first touch. Under an hour is excellent. Under a day is fine. Over a week and you are effectively doing cold outbound with extra steps.

3. Build for the evaluation maze, not the discovery phase

Since evaluation is now the longest stage, your content and your sales process should over-index there. That means:

• Security documentation available without a sales conversation, because IT security review is the single biggest bottleneck (39% overall, 50% enterprise).
• A pricing page a CFO can read. Finance involvement nearly hit 50% and CFO vetoes are running at roughly half of buyers.
• A defensible ROI story with a six-month horizon, because that is what late-veto survivors now demand.
• Flexible contract terms. 70% of buyers are being pushed toward shorter contracts by the pace of AI change, and preference for outcome-based pricing has more than doubled to 23%.

4. Arm the champion, not just the prospect

The G2 data on internal AI resistance nearly doubling to 29% is a direct instruction: your champion's hardest conversation is not with you, it is with their own colleagues. Send material designed to be forwarded internally — a one-pager, a security summary, a simple cost model — rather than material designed to be read by one person on a call.

5. Get legible to the agents

You cannot instrument the chatbot, but you can influence what it says. That means clear, structured, factual public content: an accurate product page, a comparison page that does not lie, review-site presence, and documentation that is crawlable. G2's finding that review sites now edge out chatbots as the top shortlist-shaping source is a strong argument for treating review generation as a pipeline activity rather than a marketing chore.

Being absent from an AI answer is functionally identical to not being on the shortlist. There is no page two of a chatbot response.

What this means for SDR teams and GTM agencies

For SDR managers, the metric that matters is shifting from activity to signal coverage. The right question in a pipeline review is no longer "how many touches did we run" but "how many qualified signals did we detect, and how fast did we act on them". Teams that keep optimising for volume in an AI-mediated market are optimising the wrong variable.

For founders doing their own outbound, this is arguably good news. You cannot outspend an enterprise on intent data licences, but you can absolutely outpace them on responsiveness to a public signal. Speed and specificity are cheap advantages and they compound.

For GTM agencies, signal quality is becoming the product. Clients can buy volume anywhere. The differentiated offer is a documented signal taxonomy — which triggers you monitor, how you score them, how fast you act — because that is the part that survives the shift and can be shown to work.

For RevOps, there is a data-hygiene implication worth flagging. IDC also reported that 45% of enterprise AI projects are failing to deliver results, with CIOs demanding clearer ROI and governance for agentic AI. Pointing an AI agent at bad CRM data produces confident nonsense at scale. Whatever you automate downstream, the enrichment and scoring layer has to be trustworthy first.

Takeaways

• AI-mediated B2B buying is the default, not the future. 80% of B2B tech buyers use AI agents in purchasing (IDC), and 82% have sourced software recommendations from a chatbot in the last two years (G2).
• Discovery went dark; evaluation got long. Evaluation is now the longest stage of the buying journey. Your outbound should assume the buyer is further along than your CRM says.
• Classic intent data is eroding, not dying. Account-level intent still has value — review-site behaviour especially — but the observable surface is thinning as research moves inside assistants.
• Human, public, time-stamped signals are the durable layer. Tool-recommendation requests, competitor complaints, hiring posts, funding events, post engagement, profile views, and job changes all stay visible regardless of how buyers research.
• Speed is the whole game. A signal acted on within the hour is warm outbound. The same signal acted on in two weeks is cold outbound wearing a costume.
• Sell into the evaluation maze. Security docs, CFO-ready pricing, a six-month ROI story, and shorter contract options are now table stakes, not enterprise-only concessions.
• Agents research, humans decide. Only 9% of buyers would let an agent execute a purchase inside guardrails, and 2% without pre-approval. The relationship still closes the deal — you just have to earn the conversation later in the journey than you used to.

The teams that will win Q4 are not the ones sending more. They are the ones who notice, faster than anyone else, that a specific person just raised their hand in public — and who show up with a message that could only have been written for that moment.

Sources: IDC via InfoTech Lead, August 2026 · G2 2026 Buyer Behavior Report: The Evaluation Maze · Gartner, The Future of Sales · MarketScale coverage of the IDC and Gartner data