AI for Demand Gen Marketers: The Real Playbook
Buyers rank their favorite before you ever hear from them. Where AI actually helps demand gen - account research, the 95%, outbound - and where it lies.
Demand gen is the marketing seat judged on pipeline, and the honest data about how B2B pipeline actually forms should reorganize how anyone in that seat uses AI. In 6sense’s Buyer Experience Report, built on nearly 4,000 buyer responses, 94% of buying groups had already put their vendor shortlist in order of preference before first contact with any seller, and the vendor ranked first at that moment won the deal 77% of the time. First contact now lands around 61% of the way through the journey. By the time your SDR gets a reply, the deal is mostly decided.
I read that as the job description: demand gen’s real work happens before the buyer is visible, in influencing a ranking you cannot observe directly. That reframes the AI question for this role in my role-by-role series. The loud AI use case - personalizing outbound at scale - operates on the 39% of the journey that remains after buyers have mostly decided. The quiet use cases operate on the part that picks the winner. This playbook covers both, in the order of what they are worth.
The arithmetic of the role
Two numbers discipline everything else here. The first is the 77% above: be the pre-contact favorite or fight uphill. The second is older and harsher - the Ehrenberg-Bass Institute’s 95:5 rule, from Professor John Dawes’s work with the LinkedIn B2B Institute: at any given time, up to 95% of your potential buyers are not in the market at all. Companies change providers roughly every five years, so about 20% are in market in a year and about 5% in a quarter.
Put together: most of your audience is not buying now, and the ones who are have mostly chosen before they talk to you. So demand gen splits into two different jobs - being remembered by the 95% until they enter the market, and being found and validated by the 5% while they research. AI helps with both. It helps them differently, and mixing up which job a tactic serves is where I see most AI-assisted demand gen budgets leak.
Job one: the 95%, and what AI does for memory
For the out-of-market majority, the work is presence: content, brand, and being the obvious answer when the buying trigger finally fires. AI’s contribution here is volume with a floor under quality - and one genuinely new channel.
The new channel first: buyers now research through AI assistants, which means part of “being on the shortlist” is being in the answers. That work - getting cited by AI search engines and the rest of the AI SEO discipline - is demand gen infrastructure now, not an SEO side quest. When 94% of buying groups rank vendors before contact, the sources that built the ranking did your selling, and assistants are becoming one of those sources.
On content volume: AI collapses the production cost of the steady, useful publishing that memory requires, and every workflow in my prompt library applies. The demand gen twist is discipline about what the content is for. A comparison guide is for the 5% and should be ruthlessly complete. A point-of-view piece is for the 95% and should be worth remembering. Asking one asset to do both jobs is how teams produce the beige middle thing that does neither, no matter how good the model is.
Job two: the 5%, and what AI does for finding them
Account research and prioritization. This is the highest-value AI assignment in the role, and the least glamorous. Before AI, properly researching an account - tech stack, hiring signals, recent announcements, likely initiatives - took an hour a company, so nobody did it past the top ten targets. Now the model does the assembly in minutes per account: paste what you have from your tools plus public material, get a structured brief with each claim tied to its source. My rules from the competitor analysis prompts transfer directly, especially the anti-invention constraint, because a hallucinated “signal” on an account brief sends an SDR into a call with a false premise.
Intent data synthesis. If you buy intent signals, AI is the layer that turns them from a spreadsheet nobody reads into a Monday morning brief: which accounts moved, what likely changed, what action follows. Note what stays human in that sentence. The model summarizes movement. Whether a spike is a buying committee researching or one intern writing a paper is a judgment call, and acting expensively on noise is the classic intent-data failure that AI summarization makes faster, not rarer.
Outbound personalization. The loud use case, so here is my honest position: AI-personalized outbound works exactly as long as the personalization is real. “I saw you’re hiring three RevOps roles, which usually means X” earns a reply because it is a true observation with a relevant inference. What the market is drowning in is the other kind - a scraped detail stapled to a template - and buyers pattern-matched it within months. Fake-personal is now worse than honest-brief. My rule: the model drafts from a verified account brief (see above) or it does not draft, and volume follows quality, never the reverse.
Lead scoring. Predictive scoring is the oldest “AI” in demand gen and the place to keep your skepticism freshest. A model trained on your historical conversions learns your historical go-to-market, including its blind spots - accounts you never worked score low forever, which the bias mechanics I wrote about in ad delivery reproduce quietly in pipeline form. I use scores to sequence effort, never to gate it, and I audit quarterly for whole segments the model has decided to ignore.
The weekly shape, and the stack
My honest version of the demand gen week with AI in it: Monday, the intent-and-account brief (model-assembled, human-judged) sets the week’s target list. Midweek, production - campaign assets, outbound drafts from real briefs, the 95%-facing content moving forward - with the standing context setup holding ICP, voice and claims rules so nothing starts from a blank prompt. Friday, the numbers: pipeline influenced, not activities completed, with the attribution honesty that job demands, because demand gen’s best work is exactly the influence attribution cannot see.
The stack stays smaller than the martech map suggests: one frontier assistant with standing context, your existing automation platform, whatever intent source you already pay for, and the paid channels run with the PPC manager’s audit discipline. Salesforce’s State of Marketing, across 4,450 marketing decision makers, has AI adoption at 75% - the tools are table stakes now, and the differentiation left is judgment about targets and offers, which no subscription sells.
What stays yours
The offer. AI can package and distribute an offer. It cannot know that your market is sick of webinars and would show up for a blunt teardown of their current stack. That is market feel, built from calls and communities the model is not in. (And when the right offer turns out to be a real room, the event marketer playbook covers that channel end to end.)
The sales relationship. Demand gen lives or dies on the handoff, and the handoff is politics: what counts as qualified, who follows up how fast, whose number it is when the quarter misses. An AI-generated meeting count will not save a broken agreement with sales. I would spend the hours AI returns to me on exactly this.
The definition of demand. The role’s oldest failure mode is optimizing lead volume while pipeline starves, and AI accelerates whatever you point it at. Point it at MQLs and you will drown in cheap ones. The choice of what the machine maximizes is the strategy, and it is the one input it cannot generate.
FAQ
What should a demand gen marketer automate first? Account research briefs - the highest return and lowest risk of anything I have handed to a model in this role. Every downstream tactic - outbound, ads, sales enablement - improves when it runs on verified account context instead of a company name and a guess.
Does AI outbound actually book meetings? Drafted from real research, yes, modestly - it removes the production bottleneck on genuinely personal notes. Deployed as scaled fake-familiarity, it burns addresses and domain reputation for replies that were never coming. The model is an amplifier, and it amplifies lazy exactly as efficiently as sharp.
How is this different from the growth marketer’s playbook? Motion. Growth runs an experiment loop on a funnel it can measure end to end. Demand gen works longer cycles, buying committees of ten-plus people, and influence that lands before measurement starts - which is why its AI leverage sits in research and memory-building rather than test velocity.
Will AI replace demand gen marketers? It replaces the assembly: list building, first-draft campaigns, research collation. What it cannot replace is being right about who to target and what will move them - and the buyer data above says those decisions are made further from your dashboards than ever. The role concentrates into its judgment layer. Bring judgment.
This post is one of the role-by-role playbooks - the adjacent seats are growth marketing for the experimentation loop and marketing ops for the pipes this role depends on. The buyer-side shift that makes AI answers part of demand gen is in the citation playbook, and the wider practice starts at the complete guide to AI marketing.