Your click didn’t get more expensive. It got less persuasive.

Start with the number everyone quotes. WordStream’s 2026 benchmark study reads 13,474 US search campaigns run between April 1, 2025, and March 31, 2026. Its tables say “average”; its methodology note says those are technically medians — the middle of the pack, not the mean, so a handful of monsters cannot drag the line. Across all industries: 6.64% click-through, $5.42 a click, 8.18% conversion, $66.69 a lead.

Then the line nobody put in a slide: in 2026, cost per lead went down for the first time in five years — the first decrease WordStream has measured since before 2020.

Sit with that. Leads got cheaper on Google search during a year everybody spent complaining about rising costs. The long arc is still ugly — a click cost $2.32 in 2016 and $5.42 now — but this year it eased.

So why does it not feel like it?

Because if you sell services, the money did not leak out of the auction. Business Services in the same study: $5.87 a click, 6.10% click-through, 4.85% conversion, $93.69 a lead — and a conversion rate down 5.65% year over year. The traffic arrived. It cost about what it costs. Fewer of them did the thing.

That is a different problem with a different fix, and the fix is not in the ad account.

Which is where most teams stall, because there is no shortage of things to change and no obvious order to change them in. There is an order. ConversionTeam published 2,288 A/B tests across 71 clients in June 2026, and the useful part is not the winners. It is the losers: only 19.1% of the tests reached statistical significance. Four in five changes decided nothing at all. For scale, they set that against Optimizely’s published 12% and VWO’s roughly 14%.

If four in five changes are going to tell you nothing, the only rational move is to spend your handful of attempts on the categories that win most often. By raw win rate, in their data:

Lead generation also beats ecommerce overall, 60.8% to 49.9% — a services business has better odds on any given change than a store does. Use them.

None of this says “write shorter copy” or “add testimonials,” which is what the finding usually gets flattened into. It says the sentence is the highest-leverage object on the page, the proof is second, and everything anyone was going to argue about in a design review is third. Most teams have that order backwards, because layout is the thing everyone in the room feels qualified to have an opinion about.

One honest caveat, because the data cuts both ways more often than newsletters admit. Wistia’s read of more than 13 million videos says shorter video holds attention — under a minute averages a 52% engagement rate, and it climbs the shorter you go. It says the opposite about clicks: the longer the video, the more likely a viewer is to act, with 30-to-60-minute videos taking the highest click-through. Both are true. They are two different jobs, and one asset cannot hold both. Decide which one you are making before you write the script, not in the edit.

The through-line for the rest of 2026: this is not a year to out-bid anyone. The click is priced. The persuasion is not.

The Weekly Media Review. Paste in one week of account numbers and your industry’s published benchmark. It finds the single worst step in your funnel, ranks at most three changes, and writes the kill criteria before you spend anything — which is the part everybody skips, and the reason four in five changes go on living past their usefulness.

{
  "name": "Weekly Media Review",
  "version": "1.0",
  "role": "You are a paid-media buyer reviewing one advertising account for one week. You are blunt, numerate, and you say plainly when the data cannot answer the question asked of it.",
  "inputs": {
    "business": "<what you sell, to whom, and the offer, in one sentence>",
    "industry_benchmark": {
      "source": "<name the published study and the row you are using>",
      "conversion_rate_pct": null,
      "cost_per_click_usd": null,
      "cost_per_lead_usd": null
    },
    "this_week": {
      "spend_usd": null,
      "impressions": null,
      "clicks": null,
      "landing_page_views": null,
      "leads": null,
      "qualified_leads": null,
      "closed_won": null,
      "revenue_usd": null
    },
    "prior_4_week_average": "<the same eight keys as this_week>",
    "changes_made_last_week": [
      "<one line each: creative, budget, audience, bid, page>"
    ],
    "constraints": "<budget floor and ceiling, what cannot be changed, approval lead time>"
  },
  "method": [
    "Compute CTR, CVR, CPC, CPL and cost per qualified lead for this_week and for prior_4_week_average. Show the arithmetic, not just the result.",
    "Walk the funnel step by step: impression to click, click to landing-page view, view to lead, lead to qualified lead. Name the ONE step with the largest week-over-week deterioration.",
    "Compare only that step against industry_benchmark. State whether the account is above, at, or below it, and by how much.",
    "Separate noise from signal: if a metric moved on fewer than 30 conversions, say so and refuse to act on it.",
    "Attribute the movement to changes_made_last_week only where the timing supports it, and say plainly where it does not."
  ],
  "output": {
    "format": "markdown",
    "sections": [
      "VERDICT - one sentence saying what actually changed this week.",
      "THE NUMBERS - a table: this week, prior 4-week average, benchmark.",
      "THE LEAK - the single worst step, with the arithmetic shown.",
      "DO THIS WEEK - at most 3 actions, ranked, each naming the metric it should move and by how much.",
      "DO NOT TOUCH - what is working, or too noisy to judge, and why.",
      "KILL CRITERIA - the number and the date at which the top action is judged a failure."
    ]
  },
  "rules": [
    "Never recommend more than three changes in one week. Beyond three you cannot attribute the result to any of them.",
    "Never cite a benchmark that was not supplied in inputs. Do not recall one from memory.",
    "If a required input is missing, list exactly what is missing and stop. Do not estimate it.",
    "Where qualified_leads is supplied, judge every recommendation on cost per qualified lead, never on cost per lead.",
    "End with a confidence level - high, medium or low - and one sentence saying what would raise it."
  ]
}

That is the whole format: seven blocks, same order, Wednesdays at 10:15am ET. Replies land with a person here and get answered.