If your Search performance is wobbling and your keyword tweaks aren’t moving qualified pipeline, it’s not your imagination. Google Ads is steadily removing the levers that made keyword management work. And a lot of teams are feeling it: 53% of advertisers say managing Google Ads is harder in 2026 than it was two years ago, largely because automation has reduced manual controls (State of PPC Report 2026 via Smarter Ecommerce).
That’s the uncomfortable part. The more the platform “helps,” the less a traditional keyword system behaves like a system you can actually steer.
Frederick Vallaeys—who joined Google in 2002 and served as the first AdWords Evangelist, now at Optmyzr—has been blunt about what’s happening: the keyword, as the central control surface, is becoming optional. Not irrelevant. Optional. Different thing.
Here’s why this matters right now for B2B demand gen. AI Overviews are already changing the economics of ranking for many queries: Ahrefs estimates a 58% reduction in top-ranking organic CTR for keywords that trigger AI Overviews (Dec 2025 data, Apr 2025 update). And for position 1 specifically, Ahrefs shows CTR on AI Overview keywords falling from 0.073 in Dec 2023 to 0.016 in Dec 2025—about a 78% decline.
Less click volume upstream. Less transparency downstream. Same pipeline target. That combination forces a new operating model.
Why the keyword “contract” is breaking
The original AdWords keyword model worked because it was a clean contract. Advertisers did the work (research, segmentation, match types). Google did the routing (show ads when queries matched). Then advertisers diagnosed performance in search terms reports and tuned bids, match types, and negatives.
That diagnosability is the whole point. When something broke, the system told you where to look.
But Google’s product direction has been consistent: move from explicit matching to intent inference. Google VP Dan Taylor has described the shift toward context-based matching and conversational search, saying AI matching via context reduced irrelevant ads by about 40% (via Marketing Tech News, ~2026). The implication is clear: the platform believes it can do better than static keyword patterns.
Then the platform stacked automation on top of that matching: Smart Bidding, Responsive Search Ads, Performance Max. Embryo estimates 80% of Google Ads spend is optimized by AI across those systems (stats 2023–2026). Whether the exact number is 80% or 70% in a given account, the direction is what matters: the auction is being managed by models, not spreadsheets.
So the keyword contract breaks in two places. Matching gets fuzzier. And bidding gets abstracted away from the query.
What keyword obsolescence looks like in real account math
Vallaeys’ framing is practical: keywords aren’t “dead,” but they’re turning into inputs for something else—what he calls “synthetic keywords” that represent richer intent than a literal string match. That’s a fancy label for a simple reality: the system is trying to understand meaning, not characters.
But the day-to-day pain is more basic: you’re asked to trust outcomes you can’t fully explain. Optmyzr’s PPC Town Hall has also called out reduced transparency, including less search term visibility, as a structural limitation that weakens keyword-centric optimization (Amy Hebdon, Optmyzr PPC Town Hall, 2026). When the query data is incomplete, “keyword control” becomes a story teams tell themselves.
And the spend distribution is shifting under the hood. In Optmyzr’s 2026 match type study, Exact Match share of non-branded spend dropped from 37.1% in 2022 to 27.6% in 2023. Broad Match represented 38.8% of non-branded spend and became the largest category. Phrase Match drove 40% of non-branded conversions with a 15.7% conversion rate. Meanwhile, Exact Match on branded terms yielded 6.61x ROAS at a $0.90 CPC.
That set of numbers is the whole story in miniature: protect brand (it prints), and accept that non-brand is increasingly model-driven, not match-type-sculpted.
But there’s another way to read it. The “keyword is obsolete” claim isn’t mainly about targeting. It’s about measurement. If the platform is optimizing toward conversions (and Improvado cites Google-reported conversion lifts of 14–18% attributed to AI, 2026), the question becomes: conversions of what quality, and at what unit economics?
Platform CPA can improve while qualified pipeline gets worse. Not because the model is evil. Because it will optimize to the conversion signal you feed it.
One move: replace keyword control with revenue-signal control
If you only change one thing, change this: stop treating keywords as the primary steering wheel. Treat them as guardrails. The steering wheel is your first-party revenue signal.
Here’s the 5-minute version you can run this week: build a “revenue-aligned conversion stack” so Smart Bidding (or whatever automation you’re using) has a chance to optimize for downstream value instead of cheap form fills.
The hypothesis (make it falsifiable): If we optimize Google Ads to a later-stage, CRM-confirmed conversion (or a weighted conversion set), then qualified pipeline rate will increase and wasted spend will decrease because the bidding model will learn from higher-quality outcome signals instead of top-of-funnel noise.
What to measure (and what not to over-interpret): don’t treat platform-reported ROAS as causality. Use it as directional. The north star is incremental qualified pipeline and, when possible, closed-won contribution with a lag.
Success = lift in qualified pipeline per $ (or qualified pipeline rate on paid-search influenced opportunities). Guardrails = volume (leads or MQLs) and CPL so Sales doesn’t get starved during learning. Stop-loss = if qualified pipeline drops materially for two consecutive weeks while spend holds flat, pause the experiment and revert to the prior conversion action set.
Trade-off (say it out loud): this will reduce volume before it improves quality. Smart Bidding needs time and enough conversion volume to learn; if the account is low-volume, pushing too far down-funnel can stall delivery.
Run it this week (operator-ready)
- Owner: Demand Gen lead + RevOps (shared). Paid search manager executes; RevOps validates CRM stages.
- Setup: Separate branded vs non-branded Search campaigns (Vallaeys’ recommendation). Keep brand on tighter controls; use automation more aggressively on non-brand.
- Conversion plan: Build two conversion actions: (1) “Primary” = later-stage CRM event (SQL, SAO, or whatever your org trusts), (2) “Secondary” = earlier event for learning stability. Import offline conversions where possible.
- Budget range: Use existing Search budget; don’t scale spend during the test. Keep it clean.
- Timeline: 2–4 weeks minimum for learning, longer if sales cycles are long. Read out weekly, but don’t overreact to two-day swings.
- Measurement: Use a holdout where feasible (geo split or campaign split) to estimate incrementality; if not feasible, at least compare against a baseline period with stable spend and stable conversion definitions.
When this is wrong: if the business has extremely low CRM conversion volume (or unreliable stage hygiene), pushing optimization down-funnel can make delivery worse and reporting meaningless. In that case, the first fix isn’t bidding. It’s RevOps hygiene and a tighter definition of what “qualified” means.
The kicker: keywords aren’t dying—your old job is
The keyword system used to reward craft: tight match types, meticulous structures, endless query mining. That work taught teams how buyers think because the data was visible and the controls were real.
In 2026, the scarce skill is different. It’s not picking the perfect keyword. It’s designing the signal that tells the machine what a good customer looks like—and proving, with holdouts and clean baselines, whether it actually created incremental qualified pipeline.
That’s the circle closing: as clicks get harder to win (Ahrefs) and queries get more conversational (Dan Taylor), the keyword stops being the contract. The business outcome becomes the contract. And the teams that treat measurement as a product—owned, versioned, audited—will be the ones still in control.
