Gareth Puckey Say hello

Writing new ads from what already worked in your account

Google retired the Best, Good and Low performance label. How to read asset reports and build an ad copy library from lines that worked, with CAP checks.

By Gareth Puckey Updated 7 min read Google Ads

In this guide
  1. Why the copy you need is already in your account
  2. Where to find the evidence
  3. Reading the numbers properly
  4. Building the library
  5. Carrying winners to Meta
  6. Keeping it legal: the CAP checks
  7. A monthly refresh routine
  8. FAQ

Google has retired the Best, Good and Low performance label on Search ad assets. In its place you get full statistics for each asset (impressions, clicks, cost and conversions) covering dates on or after 5 June 2025 (Google Ads Help). That makes your own account a strong source of ad copy, as long as you read the numbers carefully and keep evidence for every claim you reuse.

Why the copy you need is already in your account

Every responsive search ad (RSA) is a running test. You give Google 3 to 15 headlines of up to 30 characters and 2 to 4 descriptions of up to 90 characters, and it shows up to 3 headlines at a time (Google Ads Help). Starting each new ad from a blank page ignores the record that builds up.

A copy library should draw on lines that have already performed across accounts, refreshed every month. The library turns that data into copy someone can use.

Where to find the evidence

What replaced the performance label

The performance label ranked each asset without showing the numbers behind it. Google has deprecated it and replaced it with full statistics: impressions, clicks, cost and conversions, for dates on or after 5 June 2025 (Google Ads Help; Google Ads Help).

The asset report

The campaign-level asset report lists assets that had impressions in the last 30 days (Google Ads Help). Three things in it matter for a library:

  • The Pinned column shows which assets are fixed to a position, which affects how often they could show.
  • The Added by column marks assets Google wrote or changed as “Google AI”. Those aren’t purely your lines, so check them before copying them anywhere.
  • Google suggests replacing assets with no impressions for several weeks (Google Ads Help). A line that never serves gives you no evidence either way.

The combinations report

The combinations report shows impressions and the most common combinations of headlines and descriptions. It doesn’t show conversions, and Google says not to build static ads from it (Google Ads Help). I’d suggest using it only to see which lines tend to appear together.

From September 2026, campaigns using text customisation (formerly automatically created assets) are upgraded to AI Max automatically, and combinations reports flag combinations that include Google’s text (Google Ads Help). That flag helps keep Google’s text out of your library.

Reading the numbers properly

Treat per-asset ratios as a hint

Asset metrics are counted each time an asset serves, so the totals across assets won’t add up to the ad’s total. Per-asset click-through rate (CTR), CPA and ROAS are directional only, and Google advises judging performance at campaign or asset-group level (Google Ads Help; Google Ads Help).

A common mistake is treating one headline’s CPA from one good fortnight as gospel, when the library should hold lines that keep doing well across several ad groups over a few months. If your conversions are leads rather than sales, the same caution applies one level up, as covered in why your best campaign by leads might be your worst by sales.

Don’t chase an Excellent Ad Strength rating

Ad Strength rates ads as Incomplete, Poor, Average, Good or Excellent. It isn’t used in the auction: it doesn’t feed Ad Rank or Quality Score, and it doesn’t decide whether an ad serves (Google Ads Help).

Google does say advertisers who moved from Poor to Excellent saw 15% more conversions on average. That’s Google’s own correlational data, and Ad Strength doesn’t judge your offer or your landing page (Google Ads Help). I’d treat it as a prompt to add variety and keep proven lines even if the rating dips.

Building the library

What each entry holds

Each entry is one line, with enough context to judge it and enough evidence to run it again. Tag it with its type and character count so it drops straight into an RSA, and record impressions, period and where it ran, because a ratio without volume and dates tells you very little. This is the structure I’d suggest; the figures are made-up placeholders to show the format.

Line Type Impressions Period Seen in Evidence held Status
Free delivery over £50 Headline (22 chars) 48,000 (example) Jun to Sep 4 ad groups, 2 accounts Delivery terms confirming the threshold and conditions Ready to reuse
Prices from £49, VAT included Headline (29 chars) 31,500 (example) Jul to Sep 3 ad groups Price list showing what costs £49, with no compulsory fees added Ready to reuse
Quotes back within 24 hours Headline (27 chars) 22,000 (example) Jun to Aug 2 ad groups None yet Hold until response-time records are on file
Speak to a real person, Monday to Friday. Book a call online in a few minutes. Description (78 chars) 19,000 (example) Jul to Sep 3 ad groups Opening hours and booking page Ready to reuse
Trusted by thousands Headline, Added by: Google AI 9,000 (example) Aug to Sep 1 ad group None Do not reuse

As the last row shows, a line Google added can perform well and still be one you can’t support.

When to pin

Headline positions 1 and 2 and description position 1 always show. Headline 3 and description 2 may not (Google Ads Help). Pinning cuts the number of combinations and can lower Ad Strength, so if you pin, Google suggests pinning 2 or 3 options to each position (Google Ads Help).

I’d only pin where a line genuinely has to appear in a given spot, and leave everything else unpinned so the ad keeps generating evidence.

Carrying winners to Meta

Treat a line that did well in search as something to test on Meta, since the audience and format are different. Meta’s A/B testing guidance (Meta) sets out how:

  • Change one variable at a time, starting with creative.
  • Run the test for at least 2 weeks and no more than 30 days.
  • Expect results after about 100 events, and wait for the final result before acting.
  • Don’t use the test audience in another live campaign at the same time.

Meta claims winning tests cut cost per result by 30% on average (Meta). That figure comes from Meta itself, so judge your own tests on your own numbers.

Paid search ads fall under the CAP Code (ASA); for social, see the ASA’s social media remit page. These rules come from section 3 of the code (ASA):

  • 3.1: ads must not materially mislead.
  • 3.7: you must hold documentary evidence before the ad runs.
  • 3.18: quoted prices must include non-optional fees.
  • 3.22: “up to” and “from” claims must not exaggerate.
  • 3.23: “free” claims come with strict conditions.
  • 3.33, 3.34 and 3.38: comparisons must be like-for-like, verifiable and on a clear basis.

This is where copying “proven” lines goes wrong. “Lowest prices” may have been true when someone wrote it, and a “free” offer may have ended. Performance data says nothing about whether you can back a claim today. Keep the evidence next to each entry, and don’t reuse a line with no evidence on file. I’d check any “Google AI” assets against the same rules as lines you wrote yourself.

A monthly refresh routine

The monthly routine I’d suggest:

  1. Pull the asset report for each campaign and note which lines had impressions in the period.
  2. Set aside anything marked “Google AI” for separate review.
  3. Promote a line only when it has meaningful impressions in several ad groups across more than one period, and the campaign or asset-group results held up while it ran.
  4. Update entries with impressions, period and where they ran.
  5. Check the evidence for every claim, and move anything unsupported or out of date to “hold”.
  6. Flag lines with no impressions for several weeks for replacement, and pick one or two to test on Meta.

I use AI tools for repetitive work like this, with a person checking the output before anything goes near an account. That split is the approach in using AI in a marketing workflow without losing control, and the account checks behind it are covered in five Google Ads automations that pay for themselves.

FAQ

What replaced the Best, Good and Low performance label in Google Ads?

Full asset statistics: impressions, clicks, cost and conversions, for dates on or after 5 June 2025 (Google Ads Help).

Does Ad Strength affect Ad Rank or Quality Score?

No. Google says Ad Strength isn’t used in the auction and doesn’t feed Ad Rank or Quality Score (Google Ads Help).

How long should a Meta A/B test run?

At least 2 weeks and no more than 30 days, with results expected after about 100 events (Meta).

For the wider picture, see my guide to how Google Ads works for lead generation.

11 sources, all opened and checked
  1. About responsive search ads, Google Ads Help
  2. About Ad Strength, Google Ads Help
  3. Asset performance reporting, Google Ads Help
  4. Campaign-level asset report, Google Ads Help
  5. Asset-level metrics, Google Ads Help
  6. Combinations report, Google Ads Help
  7. Text customisation and AI Max upgrade, Google Ads Help
  8. A/B testing, Meta for Business
  9. Remit: search engines and price comparison websites, ASA
  10. Remit: social media, ASA
  11. CAP Code section 3: misleading advertising, ASA

More on Google Ads