Responsive search ads have been the standard text format in Search campaigns for years, yet most of the ones we open in audits share the same flaw: fifteen headlines saying the same thing fifteen ways. The system combines, rotates and measures, but when the material is uniform there is nothing to learn. The outcome is an ad rated “Excellent” with a click-through rate identical to one written in five minutes.
This guide starts from the mechanics, works through the three levers you genuinely control — what you write, what you pin, what you measure — and closes with the workflow we use to test copy at a time when, with AI Max, Google writes part of the ad too.
How an RSA works, minus the myths
A responsive search ad takes up to 15 headlines of 30 characters and up to 4 descriptions of 90 characters, plus two 15-character path fields for the display URL. At each auction the system assembles an ad with at most 3 headlines and 2 descriptions, choosing the combination it predicts will suit that search, that device and that context. The third headline and the second description do not always show: it depends on available space.
Two practical consequences that are routinely ignored:
- Every headline must stand on its own. You do not know which headline will appear next to which, nor in what order. A headline that only makes sense as a follow-on to another is a bad headline.
- Description 1 does most of the work. It is the one seen most often. If a message has to land, do not entrust it to the description that may not appear.
Up to three enabled RSAs per ad group are allowed, but in most accounts one or two well-built ones are enough: more ads split impressions and lengthen the wait before any reading is reliable.
Writing real variety, not paraphrase
The system can only choose among the options you give it. If your headlines are “Google Ads Consultancy”, “Expert Google Ads Consultants” and “Your Google Ads Consultancy”, you have three assets and one idea. The simplest way out is to write by message category, not by quantity.
| Category | What it answers | Roughly how many headlines |
|---|---|---|
| Relevance to the search | “Am I in the right place?” | 3-4, carrying the ad group’s core keywords |
| Value proposition | “Why you and not the others?” | 3-4, each with a different argument |
| Proof | “Why should I believe it?” | 2-3: verifiable numbers, certifications, years in business |
| Offer and terms | “What do I get, and at what price?” | 2-3: price, delivery, trial, guarantee |
| Call to action | “What do I do now?” | 2: different actions, not synonyms |
A quick check: take three headlines at random and read them as though they were the only ad you saw. If you cannot tell the ad apart from a competitor’s, the headlines are generic. If two of the three say the same thing, they are redundant.
Different lengths, different jobs
Not every headline needs all 30 characters. A short one (“Quote in 24 hours”) combines more easily and leaves room for a long, specific one. The same goes for descriptions: one expanding the value proposition, one centred on the offer, one answering the main objection, one carrying proof. Four descriptions with four jobs, not four rewordings.
A common mistake. Putting the keyword into all 15 headlines to “improve relevance”. Relevance comes from 3-4 headlines aligned with the search; the other 11 are there to make people choose your ad, not to echo the query. We made the same point in the guide to Quality Score: ad relevance is won on the ad group’s theme, not through repetition.
When pinning helps and when it hurts
Pinning fixes a headline or description to a specific position. It is a legitimate tool, but it is nearly always used for the wrong reason: the feeling of control.
When it makes sense
- Legal or regulatory constraints. A mandatory disclaimer, a financial product disclosure, wording that must always appear.
- Brand constraints. The brand name must sit in position 1 in every combination, as a documented company decision.
- A message whose absence would make the ad misleading. For instance an offer condition without which the promotional headline promises something untrue.
When it hurts
Pinning a single headline to position 1 means that position never rotates: the system only has combinations to work with in positions 2 and 3. Pinning everything, headlines and descriptions alike, turns an RSA into an old expanded text ad with extra paperwork. The system has nothing left to optimise, and Ad Strength drops accordingly — for once, rightly.
The compromise we use. When position 1 needs a constraint, we pin two or three different headlines to the same position: all of them satisfy the constraint (say, they all contain the brand), but the system can still choose which to show. It is almost always possible, and it keeps rotation where it matters most.
What Ad Strength means, and what it does not
Ad Strength — Poor, Average, Good, Excellent — is a best-practice check: it looks at how many assets you have entered, how different they are from one another, whether they include your core keywords, and how much you pin. Google says so explicitly: the rating does not directly influence the ad’s eligibility to serve. It is not part of Ad Rank and it is not part of Quality Score.
That does not mean you should ignore it. It means you should read it for what it is: an automated checklist at the writing stage, not a verdict on performance. The typical cases where it is the wrong signal:
- An ad pinned for legal reasons will score lower than one without constraints. Removing the pins to “level up” is a mistake.
- An “Average” ad can have the best CPA in the ad group. When that happens, CPA is the metric that decides, not the meter.
- “Excellent” is reachable with mediocre assets, provided there are many and they differ in wording. Different in wording is not different in meaning.
The working rule: aim for “Good” as a minimum hygiene threshold, and from there let conversions decide.
Reading the asset and combinations reports
For years the only feedback on individual assets was the performance labels (Best, Good, Low), useful for orientation and little else. Since 2025 the asset report has shown impressions, clicks and conversions for each headline and description. It is a genuine step forward, provided you understand how they are credited.
Metrics are credited to every asset present in an impression. If an ad shows three headlines and two descriptions, that single impression — and any click, and any conversion — is counted for all five assets. The consequences:
- Do not add the columns up. The sum of asset conversions is not the ad’s total conversions.
- An asset can “win” by association. A headline often shown beside a very strong headline inherits part of the credit.
- Impressions tell you first what the system is choosing. A headline with very few impressions after several weeks is one the system has already discarded: it is a candidate for replacement before you even look at CTR.
The combinations report completes the picture: it shows which sets of headlines and descriptions actually served, and how often. It is the fastest way to see the ad people really read, and it often reveals that the message you spent the most time on rarely appears.
A worked reading. Two value-proposition headlines: A with 12,000 impressions and a 6% click-to-conversion rate, B with 900 impressions and 8%. B is not “better”: with 900 impressions and a few dozen clicks the gap is well within chance, and the system is already showing it prefers A. The right move is to replace B with a different idea, not to pin it.
RSAs and AI Max: text customisation and URL expansion
With AI Max for Search campaigns the boundary of the ad shifts. Text customisation generates additional headlines and descriptions, combined with the ones you wrote, drawing on your domain, landing page, existing ads and the ad group’s keywords. Final URL expansion can choose a landing page other than the one you set, and it requires text customisation to be on: if the page changes, assets written for the original page may no longer be relevant.
What this changes for whoever writes the copy:
- Your assets become the raw material. The system uses them to infer tone and arguments. Generic assets produce generic variants.
- The website becomes copy. An expired promotion left on a page, a claim no longer approved, an out-of-date price: all of it is potential material. Before enabling AI Max, clean up the pages, not the ads.
- Text guidelines are the control. Term exclusions (words or phrases never to use) and messaging restrictions in natural language (for example “do not mention prices” or “do not promise guaranteed results”). Write them before switching on, not after.
- The asset report distinguishes the source. Review generated assets regularly: remove any that breach brand rules, and turn the reason into a new restriction.
If final URL expansion sends traffic to other pages, even the best copy performs only as well as the page it lands on — everything in our piece on landing page experience applies.
A workflow for testing copy
Copy testing with RSAs is no longer a comparison between two ads identical bar one line. It is a data-led replacement cycle, and it works if you respect the timescales.
- Write by category. 12-15 headlines spread across the five categories, 4 descriptions with four jobs. Pin only where there is a real constraint.
- Wait for volume, not days. Before judging an asset you need thousands of impressions on the ad group and a few dozen conversions. In small ad groups, reason at campaign level or consolidate.
- Read impressions first, then conversions. Assets the system hardly ever picks are replaced first. Then, among assets with comparable volume, look at CTR and conversion rate.
- Replace no more than 2-3 assets per cycle. Changing half the ad wipes the reading: you will not know which edit produced the effect.
- Replace with a new idea, not a paraphrase. If an argument is not working, try a different argument from the same category.
- For big questions, run an experiment. If you want to know whether a different offer or positioning changes the outcome, an experiment with a traffic split answers better than the asset report.
- Keep a log. Date, assets replaced, reason. Six months on, it is the only memory of what has already been tried.
This is the work we do in our free audits: before writing new ads we look at which assets the system is already discarding, and why. You can see how we work on Search campaigns, and the answers to the questions we get most often in the site’s FAQ.