Broad Match in 2026: Powerful, Cheap, and Unpoliced

Broad match spent fifteen years as the match type professionals turned off first, and Google spent the last five rebuilding it into the one they recommend most. Both eras are real. What decides which one you experience is not the algorithm — it is whether three conditions hold in your account, and the third one is the habit almost everyone skips.

How broad match changed since its reputation was earned

The broad match that earned the money-pit reputation matched on words: it saw tokens, stretched them into "related" queries, and applied your one bid to all of them. The current version matches on meaning and — more importantly — prices per query. Under Smart Bidding, every auction is bid individually using contextual signals, so a marginal match is not bought at your average bid; it is bought cheap or skipped. Matching also respects structure: a query identical to one of your exact keywords routes through that exact keyword, so broad expansion cannot cannibalise the traffic you have deliberately claimed.

Google's confidence shows up in its defaults — broad is the default match type — and in its published claims: phrase-to-broad migrations under target CPA average roughly 25% more conversions, per Google's own guidance, with a smaller conversion-value lift under target ROAS. Treat those as the vendor's numbers for the well-fed case, not a promise about your account. The honest summary: modern broad match is a genuinely different mechanism with the same failure mode, waiting for the conditions that hold it back to lapse.

The three-legged stool: signals, bidding, policing

• Leg one — conversion signals. Semantic matching optimises toward whatever your conversion tracking says success is. Thin or polluted data means expansion learns from noise. If the campaign's conversions are sparse, aggregate value upstream (account structure, conversion actions) before feeding broad match.
• Leg two — Smart Bidding. The per-query pricing described above is the safety mechanism. Broad match on manual CPC removes it and reverts the mechanism to the 2010 version.
• Leg three — query policing. Someone reads what actually matched, on a schedule, and feeds the negative list. Legs one and two are settings; leg three is labour — which is why it is the leg that fails, and the one an AI agent changes the economics of.

Where broad match wins today, and where it still loses

Wins: campaigns with steady conversion volume hunting incremental demand — non-brand prospecting, category terms, long-tail service queries where enumerating every phrasing is hopeless. The case has strengthened for a reason outside the account: search phrasing is drifting longer and more conversational as AI surfaces train people to write full questions, and enumerated keyword lists cover less of that drift every quarter. The pattern in accounts that do it well is deliberate: exact match holds the proven core, broad match runs the frontier, and monthly harvesting promotes frontier winners into exact.

Still loses: brand campaigns (where expansion mostly buys queries you would win anyway or drags in competitor traffic), compliance-heavy categories where a stray query is a policy risk, tiny budgets whose data cannot feed the learner, and any account where nobody owns leg three. Match types remain a portfolio decision, not an ideology.

A useful self-test before any migration: open the search terms report and check the date of the last negative keyword added. If the answer is measured in months, the account currently lacks leg three, and enabling broad match would be installing a bigger engine in a car with no brakes fitted. Fix the habit first — the AI-assisted version below makes that a one-week fix rather than a hiring decision.

The weekly policing prompt that makes it safe

The leg-three chore, delegated:

Pull search terms for the broad match ad groups, last 7 days. Cluster queries by theme. For each cluster: spend, conversions, cost per conversion versus the ad group's average. Flag clusters with material spend and no conversions, and any query that looks like competitor brand, careers, support or DIY intent. Draft negatives at the level you recommend — keyword, ad group or shared list — and draft any winners worth promoting to exact. Wait for my approval on both lists.

Ten minutes of judging replaces the hour of pivoting that used to make this review skippable, and the saved prompt makes week eight identical to week one instead of gradually optional. The search terms report stops being a place waste hides and becomes the instrument panel it was always supposed to be — and every negative and promotion still lands through your approval click, recorded in the audit trail.

Broad match vs AI Max: overlapping magic, different controls

Both expand you beyond your keyword list; they are not the same lever.

	Broad match	AI Max expansion
Scope	Per keyword	Campaign-level suite
What expands	Query matching	Matching + headlines + final URLs
Your controls	Match types, negatives	Text guidelines, URL exclusions, negatives
Adoption	Keyword by keyword	Toggle per campaign
Fate of the old campaign-level broad setting	—	Folds into AI Max, September 2026

The practical read: broad match is the incremental, keyword-governed way to buy expansion; AI Max is the bundled way, trading more control for more surface. Running both means two expansion engines feeding one search terms report — legitimate, but double the reason leg three exists. The September 2026 fold-in of the campaign-level broad match setting is Google signalling where the bundle is headed.

A staged migration from exact-heavy structures

Migrate like an experiment, not a conviction:

1. Verify legs one and two — trustworthy conversions, Smart Bidding stable, targets realistic.
2. Pick one campaign with volume and forgiving economics. Duplicate its top phrase or exact keywords to broad in place — do not delete the originals; identical-query priority protects them.
3. Pre-load the fence: existing negative lists attached, brand and known-junk themes excluded before the first expanded auction.
4. Police weekly with the prompt above; promote winners to exact, negate losers, and watch cost per conversion against the pre-migration baseline.
5. Verdict at 30 days: expand to the next campaign, hold, or roll back — the duplicated keywords make rollback one paused batch.

Judging the verdict honestly at day 30

The migration verdict deserves the same discipline as any experiment, because broad match fails subtly more often than loudly. Compare against the pre-migration baseline on four axes, in order: cost per conversion (the headline), conversion volume (expansion should add, not merely re-route), the share of spend landing on queries you would never have bid — read a sample, not just the totals — and negative-list growth rate, which measures how hard the fence is working. Two traps to name in advance: conversion lag flatters the "before" period if you read too early, and harvested winners promoted to exact mid-test make broad look worse than the campaign actually is — count the promoted keywords' conversions as part of the migration's yield, because they are. A migration that broke even on cost but filled next quarter's exact-match core has succeeded; a migration that "won" on volume while quietly degrading query quality has not. The numbers decide, but only if you chose which numbers before the toggle.

Broad match in 2026 is a good deal offered to accounts that can hold up their end. The algorithm brings the reach and the pricing; you bring the signals and the policing. Two of those are settings — the third is now a prompt, which removes the last honest excuse. Unpoliced is a choice now, not a constraint. Start a free pilot and put the policing on schedule before the migration starts.