How to fix: Too many keywords per ad group

TL;DR

An ad group carrying more than about 20 keywords forces a single Responsive Search Ad set to speak to many different intents at once, which dilutes ad relevance, drags Quality Score down, and starves Smart Bidding of the clean theme signal it needs to learn [1][2]. The fix is to split the sprawling ad group along its natural intent clusters into focused ad groups of 10-20 semantically tight keywords each, rewrite the RSAs per cluster so the copy mirrors the queries, and add cross-cluster negatives so each new ad group holds its lane.

Why it matters

A Google ad group exists to give one tightly themed set of keywords one tightly themed set of ads. Google's own account-organization documentation describes the structure with the electronics-store example — a separate ad group for televisions, a separate ad group for cameras, each with keywords and ads built around that single product theme [2]. The reason is mechanical: when a query enters the auction, Google matches it to a keyword, then serves an ad from the same ad group. If the ad group contains thirty mixed-intent keywords, the only RSA set available has to cover thirty different query themes simultaneously. No single combination of headlines and descriptions can be specific to all of them, so ad relevance ratings on the component report sit at Average or Below average even when each individual keyword and each individual ad would score well in a tighter context.

Quality Score is the immediate casualty. Google states that Quality Score is built from Expected CTR, Ad relevance, and Landing page experience, all three measured at the keyword level [4]. Ad relevance is the component that breaks first in an over-stuffed ad group because it scores how closely the ad matches the keyword that triggered it — and with thirty keywords pointing at one ad set, that match cannot be tight on average. Expected CTR follows, because users do not click ads that do not echo their query language. The Ad Rank multiplier compounds the damage: a keyword at QS 4 in an over-stuffed ad group loses auction eligibility to an equally-bid competitor at QS 7+, and the clicks that do fire pay an inflated CPC to compensate for the relevance penalty.

The historical context matters here. Through 2017-2021 a popular structure was the single-keyword ad group, where each ad group held exactly one keyword in three match types. SKAGs worked because near-exact match treated the keyword as a literal anchor, so one ad group really did serve one query family. In 2021 Google merged close variants into broad / phrase / exact and made match-type behavior much more probabilistic [2], and the SKAG approach collapsed — it now produces thousands of tiny ad groups that all compete with each other for the same traffic and starve Smart Bidding of signal. The correct successor is not SKAG and not the opposite extreme of fifty keywords per ad group, but a middle band of 10-20 semantically tight keywords per ad group, themed by intent rather than by exact phrase. That band is what Google's organizational guidance now points toward, and it is what the audit rule measures against.

Smart Bidding sits downstream of all this. Maximize Conversions, Target CPA, and Target ROAS strategies use ad-group-level conversion patterns as a learning signal — the bid model needs to know that a given ad group, with its specific keyword cluster and its specific RSA set, converts at a particular rate so it can bid efficiently into that pattern [3]. When the ad group is a grab-bag of intents, the conversion rate the model sees is an average across many sub-segments, none of which are bid optimally. The campaign's apparent under-performance under Smart Bidding is then often diagnosed as a bidding problem when the root cause is structural sprawl one level above.

Why the rule's thresholds: the audit treats 20+ keywords per ad group as the warning band and 30+ as the failure band, with the rule firing as failed when more than 30% of enabled ad groups cross the upper threshold. Below 15 keywords is the healthy band where intent discipline is usually intact. Beyond 30 the ad group is functionally a campaign — too many themes, no useful Ad Strength signal, no reliable Quality Score trajectory.

How to fix

1. Open the affected ad group (Campaigns → select the campaign → Ad groups → click the ad group name) and switch to the Keywords tab. Read down the list and group the keywords mentally by intent: same product or same buyer stage or same modifier family. Note the natural fault lines on paper or in a sheet.
2. Decide the split. Each new ad group should carry 10-20 semantically tight keywords that share one intent. If you find more than three clean clusters, the original ad group was acting as a hidden campaign — consider splitting at the campaign level if budgets, geo, or bid strategy also differ per cluster, otherwise keep all new ad groups inside the existing campaign.
3. Create the new ad groups (Ad groups → New ad group → Manual setup) and move each keyword cluster into the matching ad group by selecting keywords and using Edit → Move to → the destination ad group. Keep match types as-is during the move; the goal at this stage is structural, not match-type tuning.
4. Rewrite Responsive Search Ads per new ad group so the copy mirrors the cluster's query language. Build at least three RSAs per ad group with 10-15 unique headlines each, varying angle (benefit, social proof, call to action). Avoid pinning unless legal or brand mandate — pinned positions block Smart Bidding from running the combinatorial test that justifies having multiple RSAs in the first place [1]. An acceptable pin is brand name in headline 1. See RSA pinning and Unique RSA headlines for the companion creative fixes.
5. Add cross-cluster negatives so each new ad group holds its lane (Keywords → Negative keywords → ad group level). If you split a sprawling ad group into "running shoes" and "hiking boots", add "hiking" as a negative to the running shoes ad group and "running" as a negative to the hiking boots ad group. Without these, the keywords will cross-fire and you have rebuilt the original sprawl in two ad groups.
6. Hold the structure for 7-14 days before pausing the original sprawling ad group, so the new ad groups absorb its traffic and Smart Bidding can re-stabilize. Smart Bidding re-enters a learning phase after major structural changes [3], so do not stack budget shifts beyond 20% or bid-strategy swaps during the window. Re-run the audit at the end to confirm the rule transitions from failed to passed.

Common mistakes

• Splitting by match type instead of by intent. Three ad groups for "running shoes" — one exact, one phrase, one broad — recreates the SKAG-era pattern that no longer works after the 2021 close-variant changes. Smart Bidding handles match-type signal internally; what you owe it is intent discipline, not match-type discipline.
• Shipping the split without rewriting the RSAs. Moving keywords into new ad groups while leaving the original generic RSAs attached defeats the entire fix — the relevance gain comes from RSAs that mirror the new tight theme, not from the structural move alone.
• Forgetting cross-cluster negatives. Without negatives blocking the other clusters, broad and phrase keywords cross-fire and the split collapses within a week back into a sprawl-of-two-ad-groups.
• Going to the SKAG extreme. Splitting a 30-keyword ad group into 30 single-keyword ad groups is a 2017-era pattern, not a 2026 fix. Aim for clusters of 10-20 tightly themed keywords; below 3 keywords per ad group you trigger the under-built ad group failure mode at the other end of the spectrum.
• Pausing the original ad group too quickly. Pausing on day one starves the new ad groups of the historical Quality Score and Smart Bidding signal that would have transferred via traffic continuity. Run both in parallel for 7-14 days, then pause the original once the new ad groups are absorbing its volume.

FAQ

Is 20 keywords a hard limit? No. 20 is a soft target with a warning band at 20-29 and a failure band at 30+. Some merchants run productive ad groups at 25-30 when all keywords are very tightly themed. The rule flags risk, not a hard rule.

Does this apply to PMax campaigns? No. PMax uses asset groups, not ad groups — the equivalent there is keeping each asset group focused on one product or audience theme. See the Performance Max asset completeness rule for the PMax-side baseline.

Will splitting hurt my Quality Score history? Quality Score history is keyword-level, not ad-group-level. Moving a keyword to a new ad group does not reset its history — it carries forward. The improvement comes from Ad relevance rebuilding around the tighter theme, not from a reset.

Should I worry about diluting conversion volume across more ad groups? Smart Bidding's 30-conversions-per-30-days learning gate is at the campaign or portfolio level, not the ad group level. Splitting ad groups inside one campaign does not split the bid model's signal pool [3]. Splitting at the campaign level does, so keep splits inside the existing campaign unless budgets or geo also differ.

What if I have an ad group with 100+ keywords? Audit it for whether it should be a campaign of its own — multiple budget needs, multiple bid strategies, multiple geo targets often hide inside such monsters. If it is genuinely one intent at scale, the fix is the same as for 30-keyword ad groups: cluster by intent, split into 5-10 ad groups of 10-20 keywords each.

Sources

1. Google Ads Help — About responsive search ads. Documents the RSA combinatorial test logic, the 3-15 headline range, the recommendation to run at least three RSAs per ad group, and the rationale for avoiding aggressive pinning — the canonical source for the per-ad-group RSA half of this fix.
2. Google Ads Help — Organize your account with ad groups. The Google-authoritative description of ad group structure: grouping related keywords by a single theme (product, buyer stage, modifier family) so the ad set can address that theme directly. The electronics-store example in the doc illustrates the intent-cluster pattern this article advocates.
3. Google Ads Help — About Smart Bidding. Describes how Smart Bidding strategies (Target CPA, Target ROAS, Maximize Conversions, Maximize Conversion Value) use ad-group-level signal patterns and re-enter learning after structural changes — referenced here for the 7-14 day hold-still window after the split.
4. Google Ads Help — About Quality Score for Search campaigns. The three Quality Score components (Expected CTR, Ad relevance, Landing page experience) and the mechanism by which sprawling ad groups push Ad relevance toward Below average — the diagnostic underpinning of the Why it matters section.
5. Wikipedia — Google Ads. Background reference for the Google Ads platform, ad group surface area, and the 2021 close-variant match-type changes that made the SKAG pattern obsolete.