The guide · for PMMs and marketing leads

The best things PMMs should be doing with AI in 2026

Most “AI for marketers” lists tell you to brainstorm headlines and summarize meetings. That's 2024 advice, reheated. The marketers pulling ahead in 2026 aren't using AI to write faster, they're using it to know things first: what a competitor changed, what their customers started complaining about, which deal is about to get blindsided.

Here's the honest list. Hardest-hitting first, and every item ends with something you can run today, mostly free.

1. Give your AI a job, not a prompt

Watch a PMM use their assistant and you'll usually see the same loop: type “analyze my competitor,” get five confident paragraphs, feel vaguely unsatisfied, close the tab. The problem isn't the model, it's that you handed a brilliant generalist a vague request and no method.

The fix is a skill: a small file you paste once that teaches your assistant an entire workflow, what to check, in what order, what honest sourcing looks like, what the final deliverable is, and when to admit it can't see something. After that, “catch me up on [competitor]” stops producing vibes and starts producing a structured, honestly-labeled brief. Same assistant, same subscription. The difference is that someone finally wrote the job description.

Two things change when you work this way. First, output quality stops depending on how good your prompt happened to be at 4:50pm on a Friday. Second (and this is the part nobody tells you) the skill teaches you the method as it runs. You watch it check pricing, then launches, then reviews, in order, every time, and that becomes how you think.

2. Demand receipts from every AI answer

The most dangerous thing your assistant does is answer confidently about things it cannot see. Its training data ended months ago , before your competitor's last two launches, their pricing change, their quiet enterprise pivot, and it will still answer “what does their Pro tier cost?” fluently, instantly, and maybe wrong. Fluency is not freshness. It reads as knowledge; it's memory.

So make one rule for anything that leaves your laptop in a deck, a battle card, or an exec email: every claim carries a source link and a date, or it's labeled as a guess. Not most claims. Every claim.

The cheapest way to enforce it: tell your assistant, in your standing instructions, to append “as of [date]” to any factual claim about a competitor and to say “I can't verify this” when it can't. Then watch what happens to your next competitive summary. The first sourced pass is usually a bloodbath, half of what the model “knew” turns out to be undated, and a chunk of it is stale. Better to find out in your draft than in a QBR when the CRO's slide contradicts yours.

This rule costs nothing, works with any assistant, and changes your relationship with AI output permanently. It's also the reason everything else on this list works: once you demand receipts, you find out exactly where your assistant is blind, and blindness, unlike wrongness, is fixable.

3. Run a Monday pulse on every competitor

Ask a marketer who “always knows first” how they do it and the answer is never talent. It's a standing ritual. Every Monday, before coffee: what did each tracked competitor do in the last week, launches, pricing-page changes, review-sentiment shifts, notable LinkedIn activity, key hires. Twenty minutes, same order, every week.

The old version of this ritual was eleven browser tabs, three Google Alerts digests that mostly catch press releases, and a Notion doc nobody else updates. Call it two hours, which is why almost nobody actually did it weekly. The 2026 version: your assistant runs the pulse skill, walks the same checklist, and hands you a brief, one bold takeaway per section, evidence underneath. Reading it takes eight minutes. Whether each item carries a live source link or an honest “as of my training data” label depends on what your assistant can see (item 2), but even the labeled version beats tab-hopping, because it's consistent and it never skips a week.

Do this for a month and something shifts: sales stops surprising you. That's the whole job, some weeks.

4. Mine your rival's reviews for your positioning

Your competitor's G2 page is a positioning brief written by their own customers, published in public, updated continuously, and read by almost nobody on your team. That's a gift.

• ›Cluster the complaints. Not one-by-one reading , have your assistant group them into themes and rank by frequency × recency × emotional heat. “Support is slow” appearing twelve times since March beats one furious outlier from 2024.
• ›Find the “love it, but…” ceilings. The most useful reviews are positive ones with a reservation, “great product, but reporting is basic.” That's a satisfied customer telling you exactly where the product stops.
• ›Catch switching signals. “We moved from…”, “compared it with…”, “wish it had…”: these phrases mark people mid-decision. Their language becomes your comparison page.
• ›Map every cluster to a strength you actually have. And when you don't have one, say so internally, an honest “we don't win here” saves sales from walking into it.

That's not spin. That's listening at scale, and it produces sharper positioning than another internal workshop about your own product, because it starts from their customers' words instead of your team's assumptions.

5. Treat a rival's pricing change as a 24-hour drill

Here's a Tuesday that happens at every B2B SaaS company, sooner or later. Sales pings you: “[Competitor] is undercutting us, we just lost one on price.” Within a day, that sentence hardens into internal fact. Within a week, someone proposes matching. Whether any of it is true often never gets checked.

• ›Verify what actually changed. Price level? Packaging? Usage gates? Billing terms? “They're cheaper now” usually decomposes into something narrower, an annual-commitment discount, a new entry tier, a repackaged cap.
• ›Read the intent. A price cut with a new low tier reads down-market push. A cut on annual-only reads cash-flow or retention pressure. Removed public pricing reads enterprise pivot. Each implies a different response.
• ›Map the blast radius. Which of your segments and live deals does it actually touch? Often the honest answer is “two segments, not the whole book.”
• ›Pick a play and arm sales. Hold and reframe value, targeted counter-offer, repositioning, matching is the last resort, not the reflex. Whatever you choose, sales gets one page: what changed, what it means, what to say, what not to say.

Guessing is how companies match a price nobody actually cut. The bare-assistant version of this drill can read today's pricing page and walk the interpretation frame with you, which is already better than the hallway version. What it structurally cannot do is see what the page said before or when it changed; if the drill matters to you, that's the one gap to close (see the thread at the end).

6. Build battle cards from evidence, not adjectives

“They're clunky and expensive.” That's not a battle card, that's a mood. Sales discovers the difference mid-call, exactly once, and then stops using your cards forever, which is the real reason most battle cards die: not laziness, lost trust.

• ›How to spot them in a deal, the tells: which personas bring them up, what trial behavior looks like, which feature questions signal their pitch landed first.
• ›Their pitch, and its cracks, stated in their own words, quoted from their site and materials, with links. You can't counter a pitch you've paraphrased into a strawman.
• ›Three counter-angles you can prove, each anchored to evidence: a sourced customer complaint of theirs, a dated limitation, a benchmark you can defend.
• ›Objection-handling lines, the actual sentences, because a stressed AE on a call doesn't have time to improvise from your bullet points.
• ›What NOT to say, the claims that used to be true and aren't, the feature gap they closed last quarter. One outdated attack costs the card its credibility.

The standard that makes it work is the same as item 2: if a claim has no source, it doesn't go on the card. Cards built this way age visibly, you can see the dates get stale , which means they get refreshed instead of quietly distrusted.

7. Write positioning WITH the frameworks, not from memory

You know April Dunford's positioning canvas exists. You know jobs-to-be-done exists, and the StoryBrand spine, and the Heath brothers' stickiness checklist. Knowing they exist is not the same as having time to reread four books before every launch, so most positioning gets written from a half-remembered summary of a framework someone read in 2021.

The 2026 move: frameworks as skills. The method lives in a file; your assistant walks you through it step by step, asks the canvas's questions in order, pushes back when your “differentiator” is a feature every competitor lists, refuses to skip the competitive-alternatives step you always skip. You bring judgment and context; it brings the discipline of actually following the method.

An entire ecosystem of these exists free, open-source libraries of book-framework skills you can install in minutes. Start there. The point isn't any single framework; it's that method-as-a-file beats method-as-a-memory.

8. Make your AI the first reader of everything

Before your launch page, deck, or one-liner ships, run the cheapest message test that exists: open a fresh session, one with no memory of your product, no context, nothing, paste the draft cold, and ask three questions:

• ›“What does this company sell, in your own words?”
• ›“Who is it for?”
• ›“Why would they pick it over the obvious alternative?”

If a fresh reader with superhuman reading comprehension can't play your pitch back accurately after one pass, your prospect, skimming on a phone between meetings, has no chance. You'll catch the jargon you've gone blind to, the buried lede, the three value props fighting for one headline.

This is not a substitute for showing it to real customers. It's the filter before that: it catches the failures too embarrassing to spend a customer conversation on, and it costs ninety seconds. The one rule: the session must be fresh. Your usual assistant has absorbed your internal vocabulary and will flatter you.

9. Keep score honestly

Every practice above can degenerate into playing with AI instead of working with it. The difference is a scoreboard. One metric per practice, counted weekly, nothing fancy:

• ›Pulses run per week vs competitors you claim to track (item 3).
• ›Sourced-claims ratio in your last shipped deck , claims with a link and date over total claims (items 2, 6).
• ›Flag-to-armed time, hours from “sales flagged a competitor move” to “sales has the one-pager” (item 5).
• ›Cold-read pass rate, did a fresh session play your pitch back right, first try? (item 8).

If a quarter of AI usage hasn't moved one of those numbers, the tooling isn't the problem, the ritual is. Cut what you're not actually running and double down on the one habit that moved a number. Four practices run weekly beat nine admired from a distance.