How to choose a search marketing agency that actually drives

WordStream’s mid-2026 benchmark puts average Google Search CPC near $2.96, up roughly 12% year-over-year, while AI Overviews now sit on more than 20% of queries and cut organic CTR by nearly 60% when they appear. That is the operating environment a marketing leader hands a search partner, not a slide about “being found.” Search marketing agencies exist to keep paid and organic demand contributing pipeline when clicks get scarcer, bidding gets noisier, and last-click reports lie.

If you already own the marketing number, you do not need a glossary of SEM versus SEO. You need a partner that will refuse to dump branded queries into Performance Max without negatives, will not ship paid traffic to the homepage, and will not promise a CPL before they have seen your conversion plumbing. This piece is about that operating standard: what the work actually is, where budgets still leak 20-30% into junk queries, and a checklist you can run before you renew or hire.

The 2026 job is signal quality, not another campaign rename

Google Ads still prints an average 4.2x ROAS across industries. Meta sits closer to 2.8x. Those are channel averages, not a forecast for your account. The gap between a 4x account and a leaking one is rarely “we need more keywords.” It is conversion data loss (ad blockers and privacy changes routinely wipe 20%+ of events), landing pages that do not match the query, and smart bidding trained on form fills that sales later rejects.

Paid search still needs keywords. “Signals over keywords” is real for bidding and creative, not a license to delete intent. High CPC pressure is pushing some teams back to manual bidding on new accounts because automated bidding inflated spend before quality caught up. That is a training-data problem. If Enhanced Conversions, a clean GTM container, and a conversion action that maps to a qualified lead are missing, the algorithm will happily buy cheaper junk.

Organic is equally unforgiving. 68% of US Google searches ended without a click in early 2026. A 2025 BrightEdge / Originality.ai analysis found 86%+ of top-ranking pages contain detectable AI-generated text, so undifferentiated blog volume is not a strategy. Information gain, citation share, and pages that still convert when the SERP is an Overview matter more than ranking screenshots. Mobile already takes 63%+ of paid search clicks, yet conversion rates lag desktop by 30-40%. If your agency reports blended CVR, they are hiding the mobile hole.

Google’s CrUX data through mid-2026 shows fewer than 60% of origins passing all three Core Web Vitals - check your own origin at web.dev/measure before assuming you are in the passing cohort. INP failures on quote and demo forms are a search problem, not a “dev backlog” footnote. A search marketing agency that cannot talk CWV, query mapping, and conversion actions in the same meeting is selling channels, not pipeline.

Where retainers go when query hygiene and landing match are ignored

Typical PPC fees still land around 10-20% of spend. Paid media retainers run $8K-$25K a month; fuller search-plus-landing programs sit higher. Median digital retainers hover near $3,000, which is why so many programs look cheap and perform like it. Pricing varies 5-10x for similar-sounding scopes. The question is not the invoice line. It is whether anyone owns negatives, search terms, dedicated landing pages, and weekly insight instead of a monthly PDF.

Twenty to thirty percent of SEM budgets still vanish on irrelevant queries, neglected negatives, or pages that do not continue the ad promise. Performance Max can drive the bulk of spend and still print 4x-8x ROAS in accounts with clean conversion setup and fewer, broader campaigns. It can also launder brand terms and shopping junk into a blended number you cannot defend to finance. If the partner cannot show brand versus non-brand, Search versus PMax, and qualified-lead versus raw form, you do not have a search program. You have a dashboard.

Buyers already say the quiet parts: “If they say we just send traffic to your homepage, run.” “Who specifically will manage my account?” “Guaranteed results” before anyone has logged into the account is a red flag. Realistic compounding on search is 3-6 months. Anything that promises a fixed ROAS on day one has not seen your unit economics.

Attribution is tactically useful and strategically incomplete. Last-click double-counts. CM360-style views from Google Ads have been getting thinner since late 2025. You still need in-platform diagnostics for weekly decisions. You also need a business-level answer, incrementality tests, and first-party data so bidding is not flying blind. Search marketing agencies that only defend platform CPA are not ready for a CFO conversation.

If paid search is leaking into the homepage and you want a partner that will rebuild the query-to-page map rather than add another campaign, HeyLead’s Google Ads and SEM programs are built around that join, not a media plan that stops at the click.

A working search partner: paid intent, organic coverage, and the page after the click

Treat search as one demand system. SEO covers the questions buyers ask before they will click an ad. SEM captures the queries where you cannot wait for rankings, or where AI Overviews stole the organic click. Landing pages and measurement decide whether either channel was worth the CPC. That is the full-stack partnership language buyers already use: tighter positioning, clearer offers, decent attribution, not another platform bolted on.

On paid, the 2026 pattern that holds up is fewer campaigns, broader targeting, and ruthless query control. Google’s broad-match plus Smart Bidding combination can deliver 12-18% lower CPCs versus tightly matched campaigns when conversion signals are clean - per Google’s own case studies - but the efficiency evaporates without tight negatives. They will also spend into garbage if negatives and brand exclusions are lazy. “Run broad” only works when creative, offer, and the post-click page are specific. Same logic as Meta’s “creative is your targeting,” applied to RSA assets, sitelinks, and the H1 the clicker actually sees.

On organic, stop measuring success as session volume. Median organic CTR already varies wildly by vertical. Zero-click search and AI Overviews mean impression visibility, AEO, and being the brand cited in the Overview can matter more than a vanity CTR. Pages need human perspective and proof, not another AI draft that reads like every other top result. Cadence still matters; undifferentiated cadence does not.

After the click, dedicated pages beat homepages. One operator moved landing CVR from 33.22% to 43.52% by adding proof above the form. CRO work has produced 70% lifts in PPC conversion rate when the page actually continued the ad. Use HeyLead Insights to see scroll depth, rage clicks on the form, and where mobile INP makes people bounce before the thank-you page. Guessing is how you keep buying $2.96 clicks that never become pipeline.

Run this 10-step search brief before you hire or renew

Do this on your own accounts first. You will either fix the leaks or walk into an agency conversation with evidence instead of vibes. A demand-gen lead at a B2B services firm did exactly that: 11 days of search-term exports showed 27% of spend on “jobs,” “salary,” and DIY how-tos. Negatives plus a consult landing page (not the homepage) dropped wasted spend before they touched bids. CPC barely moved. Qualified rate did.

Another operator, a VP Marketing at a mid-market SaaS company, had PMax “driving the bulk” at a blended 5.1x while non-brand Search sat at 1.8x after stripping brand. They split brand, rebuilt conversion actions to demo-booked, and stopped feeding newsletter events to bidding. ROAS on non-brand did not magically jump in a week. Qualified pipeline from search stopped being a mystery number in the CRM.

If organic coverage is the thin side of the same system, a specialist still has to ship pages that add information gain, not more AI sameness. That is slower than a PMax toggle and more durable when Overviews steal the easy click.

Patterns we see in new accounts

PMax often swallows brand and prints a blended 6x ROAS until finance asks for non-brand only. The real number is ugly when the conversion action is still a form fill sales never touched. Split brand, rebuild conversion actions to a sales-ready event, and stop feeding newsletter events to bidding so qualified pipeline from search is visible in the CRM.

Negatives are frequently the whole early win. Almost a third of Search spend can sit on career and how-to queries. CPC does not have to crash for the calendar to fill with consults instead of tyre-kickers - query hygiene plus a dedicated landing page (not the homepage) is the pattern that shows up first.

FAQs

Should search marketing agencies guarantee ROAS or CPL up front?

No. Nobody can honestly lock a CPL before they have seen your account, landing pages, close rates, and conversion quality. Use a 3-6 month window with leading indicators: wasted query share, qualified rate, and non-brand efficiency. Guarantees are a sales tactic, not a media plan.

Do we still need keywords if bidding is signal-led?

Yes. Keywords and negatives define where you are willing to show. Signals (first-party conversions, Enhanced Conversions, offline import) decide how Google bids inside that box. Dropping intent control because “AI Max is cheaper” is how 20-30% waste returns.

Is Performance Max enough on its own?

PMax can carry a lot of spend and hit 4x-8x in clean accounts. It is a poor sole source of truth. Keep a readable Search layer for non-brand intent, exclude or isolate brand, and do not let shopping or display bleed into the number you report as “search.”

How should we handle AI Overviews cutting organic CTR?

Measure impression and citation presence, not only clicks. Strengthen pages with proof and unique answers so you are the brand mentioned in the Overview. Use paid search to recapture high-intent queries where organic is zero-click. Do not buy more blog volume as a consolation prize.

Who should own the Google Ads account?

You should. The agency works inside your MCC with named access. Account lock-in is a real objection. History, negatives, and conversion setup are company assets, not agency leverage.

Putting it to work

This week

• Export 90 days of search terms and highlight spend that cannot become a qualified lead.

• Screenshot every Primary conversion action and mark which ones sales actually wants.

• List the top 10 non-brand ads and the URLs they hit; flag homepage landings.

• Split last month’s ROAS brand versus non-brand so finance sees the real search number.

• Run Core Web Vitals and index checks on those money URLs.

• Write the first negative list (jobs, DIY, out-of-geo) and ship it.

Next 30 days

• Launch or rebuild one dedicated landing page per top intent cluster.

• Turn on or repair Enhanced Conversions and UTM-to-CRM mapping.

• Freeze structure; test RSA proof and form friction, not new campaign trees.

• Agree an agency scorecard: wasted spend, qualified rate, mobile CVR, named lead.

• Revisit organic pages that show in AI Overviews and add unique proof, not filler.

Start by pulling 90 days of search terms next to qualified leads by source, not last month’s blended ROAS. If the join between query, landing page, and conversion action is the part your team cannot staff every week, HeyLead will take that search hygiene, bidding-signal, and post-click work off your plate so paid and organic keep contributing pipeline instead of blended vanity. Chat with us on WhatsApp