Nobody Is Writing Your Google Ads Any More

Somewhere in your account there may be an advert running right now containing a line that nobody in your business ever typed.

Not a headline you forgot to write. A sentence assembled from your landing page copy and your existing assets, generated to match a particular search.

That’s not a glitch, and it isn’t Google going rogue. It’s a feature; it has a name, and as of this month it’s arriving in accounts whether or not anyone asked for it.

Google Ads has been drifting from a writing job to an editing job for years. This is the point where a lot of advertisers notice.

There are two separate things happening, and most articles cover only one of them.

The first is responsive search ads, which have been assembling your adverts from parts for years. Most people understand that by now.

The second is AI Max, which is newer, and which changes what the parts are and whether your instructions about them still hold.

WHAT A RESPONSIVE SEARCH AD ACTUALLY IS

Worth a quick refresher, because the mechanics matter for everything that follows.

You don’t write an advert any more. You write components.

You supply up to fifteen headlines of thirty characters each, and up to four descriptions of ninety characters each. The minimum is three headlines and two descriptions.

Google then assembles combinations from what you gave it, usually showing two or three headlines and one or two descriptions at a time, and it decides which combination to serve for a given search.

Why that changes how you write

You aren’t writing a sentence. You’re writing parts that have to work in any order, alongside almost any other part.

That kills a lot of ordinary copywriting instincts. You can’t build to a punchline, because the punchline might appear first. You can’t rely on one headline setting up another, because they might never appear together.

Every headline has to stand alone and make sense wherever it lands.

That’s the discipline, and it’s why so many accounts end up with fifteen headlines that are really the same sentence rephrased fifteen times.

WHAT CHANGED THIS MONTH

Here’s the part with a date on it.

AI Max is a suite of features Google layers onto existing Search campaigns. It is not a new campaign type. It is an optimisation layer applied to what you already run.

Google’s components for it are search term matching, text customisation, final URL expansion, locations of interest, brand settings, URL inclusions and exclusions, and reporting updates.

Search Engine Land reported the migration timeline. From early August 2026, campaign-level broad match and legacy automatically created assets can no longer be created in the interface, Editor or API. Existing campaigns using them migrate to AI Max during September 2026.

So this isn’t a thing to consider next year. If your account uses campaign-level broad match or automatically created assets, the migration window is now.

What that means practically

If you have never heard of AI Max and you run Search campaigns, it’s worth an hour this month rather than a surprise in October.

Not because it’s necessarily bad. Plenty of accounts will do well out of it.

Because a change to how your adverts are written and where they send people is not something to discover from a performance dip six weeks later.

WHAT TEXT CUSTOMISATION DOES TO YOUR COPY

This is the feature that produces the sentence nobody wrote.

Text customisation is the new name for what used to be called automatically created assets. Google states that it uses text from your existing ads, your landing page copy and your assets, along with generative AI, to create customised ad copy that’s more relevant to specific searches.

Read that carefully. Your landing page copy is an input.

So the words on your website can become the words in your advert, recombined by a system, without passing anybody’s desk.

Where this goes well

For a business with well-written pages and a broad range of services, it can genuinely help. It produces relevance you’d struggle to hand-write across hundreds of search variations.

If your pages are good, the raw material is good.

Where it goes badly

If your landing pages contain vague marketing language, that’s what gets harvested.

If they contain claims you’d never make in an advert, or old pricing, or a service you stopped offering, those become candidates too.

And in regulated trades, where wording matters legally, generated copy is a genuine risk rather than a theoretical one.

Your website is now a source of ad copy. That’s a good reason to read it as though it were.

The action here is not necessarily to switch the feature off. It’s to go and look at what your pages actually say, because they’re feeding the machine.

THE PIN THAT STOPS WORKING

This is the detail I’d most want a business owner to know, and I have not seen it discussed anywhere on page one.

Pinning is how you keep control. You pin a headline to position one, two or three, and it appears there every time. It’s how advertisers guarantee their brand shows, or a required regulatory phrase, or a legally necessary qualifier.

Google’s own documentation on AI Max states that when final URL expansion is enabled, pinned assets are not used if a more relevant URL is chosen, because they may not match the customer’s intent. Similar restrictions apply with URL inclusions.

Why that matters more than it sounds

Every pinning guide teaches pinning as a guarantee. It isn’t one any more, under that setting.

If final URL expansion sends somebody to a different page than the one your advert was built around, the assets you pinned may simply not appear.

For most businesses, that’s a nuisance. For a business with a compliance line that has to be shown, it’s a problem worth understanding before September rather than after.

What to do about it

Check whether final URL expansion is enabled on your campaigns.

If you rely on pinning for anything that must always appear, that combination needs a deliberate decision rather than a default.

And if you have regulatory wording obligations, this is the paragraph to take to whoever manages your account.

PINNING, AND HOW TO USE IT PROPERLY

Setting AI Max aside, pinning is still the main control you have, and most accounts use it wrong in one of two directions.

Some pin nothing, and end up with adverts where the brand sometimes never appears. Others pin everything, which reduces the combinations Google can test down to almost nothing.

Google’s guidance is to avoid pinning identical or very similar text to the same slot, since that restricts the system and can lower your ad strength.

The three bucket method

The approach I use is to author headlines in three deliberate groups, and pin by group rather than individually.

Headlines one to three are the keyword bucket. These repeat the language of the search, so whichever appears, the advert matches what was typed.

Headlines four to six are the benefit bucket. What the customer actually gets. Speed, price certainty, guarantee, coverage area.

Headlines seven to nine are the call to action bucket. What to do next.

Pin each bucket to its position rather than pinning one specific line. So position one always draws from the keyword group, position two from benefits, position three from the call to action.

Why this works better than pinning individually

You keep the structure of the advert while leaving Google room to test within each slot.

The advert always reads sensibly, because a keyword line is always first and a call to action is always last. But there are still dozens of combinations rather than one fixed sentence.

Put a short pin guide at the top of your ad copy sheet so whoever edits it next understands the structure. Ad copy sheets outlive the person who wrote them.

AD STRENGTH IS NOT PERFORMANCE

Google rates every responsive search ad from poor to excellent, and businesses chase that rating far harder than it deserves.

Ad strength measures the quantity and diversity of the assets you supplied. More unique headlines, fewer near-duplicates, more descriptions, less pinning.

It’s a completeness measure. It’s not a performance measure.

What it doesn’t know

It doesn’t know your conversion rate. It doesn’t know your margins. It doesn’t know that the enquiries from one version were worth having and the enquiries from another were not.

An advert rated excellent can sell less than one rated average, and it happens regularly.

There’s also a direct tension. Pinning lowers ad strength by design, because it reduces combinations. So an advert with a properly pinned structure often rates worse than a loose one that says nothing in particular.

How much to care

Use it as a checklist rather than a target.

Poor is worth investigating, because it usually means you supplied too few assets or wrote fifteen versions of the same line.

Average with deliberate pinning is a perfectly reasonable place to sit. Excellent is not the goal, and chasing it by unpinning things you needed pinned is how businesses lose control of their own adverts.

WRITING FIFTEEN HEADLINES THAT ARE NOT THE SAME SENTENCE

The practical problem everyone hits. Three headlines are easy. Fifteen distinct ones is genuinely hard.

Here’s how I’d generate them without repeating yourself.

Start from the search, not the business

Three headlines that repeat what people actually type. Take them from your search terms report, not your imagination.

If people search “emergency electrician”, one headline should say emergency electrician. Not “24/7 Electrical Solutions”, which is what you’d write if you were describing yourself.

Then the specifics

Numbers, timescales, coverage. “Attend Within Two Hours”. “Fixed Price Before We Start”. “Covering All Of West Yorkshire”.

Specifics do more work than adjectives, and they’re easier to write differently from one another.

Then the objections

What stops somebody from calling? Cost uncertainty, trust, availability, tie-ins.

Each objection is a headline. “No Call Out Charge”. “Fully Insured And Accredited”. “No Long Contract”.

Then the proof

Years trading, accreditations, review scores, guarantee length.

Then the actions

“Get A Quote Today”. “Speak To An Engineer”. “Book A Free Survey”.

Five groups, three lines each, and you have fifteen headlines that genuinely differ rather than fifteen paraphrases.

Check them mechanically before they go in

Read every line on its own and ask whether it makes sense with no other headline beside it.

Then check the dull things. No truncation at thirty characters. No abbreviations a customer wouldn’t use. No stray characters or mangled punctuation from a spreadsheet.

Those mechanical faults are invisible in a document and very visible in a live advert.

WHAT YOU STILL CONTROL

I have spent most of this post on what has been handed over, so let me be clear about what has not. It is more than people assume.

You control the raw material

Every headline and description you write is still yours. Generated copy draws on your assets and your pages, so what you supply sets the ceiling on what can be assembled.

Write fifteen vague lines, and you get vague adverts, whoever assembles them.

You control the pages being read

Text customisation uses your landing page copy. That copy is entirely yours to fix.

Which makes reading your own pages a piece of advertising work now, not just website housekeeping.

You control the brand settings

Brand settings offer inclusions and exclusions at campaign and ad group level. That is a real lever over how your name is used and which brands you appear against.

You control what stays off

Search term matching can be toggled off in the ad group settings. Final URL expansion is a setting. These are choices rather than fixed conditions.

The point is not that you should switch everything off. Plenty of it performs. The point is that being defaulted into something is different from choosing it.

You control the measurement

Nothing in any of this decides what a good enquiry looks like in your business. That judgement is still entirely yours, and it is the one that matters most.

DESCRIPTIONS, THE PART EVERYONE IGNORES

Headlines get all the attention. Descriptions get written in a hurry and then never touched again.

You get four, at ninety characters each, and Google shows one or two at a time. Most accounts I see have written two, and one of those is a sentence about how long the company has been trading.

That is a lot of unused space.

What descriptions are actually for

Headlines grab. Descriptions answer the next question.

Somebody has read a headline saying you fit flat roofs. The question forming in their head is what it costs, how long it takes, or whether you cover their area.

That is what the description is for. Not a summary of your values.

How to write four different ones

Give each a different job, the same way you did with headlines.

One handles price or cost certainty. Fixed quotes, no call-out charge, free survey.

One handles speed or availability. Same-day, within two hours, weekend cover.

One handles trust. Accreditations, guarantee length, years trading, insurance.

One handles the action, with a clear instruction and a reason to do it now.

Four descriptions doing four jobs beats four paraphrases of the same reassurance, and it gives the system genuinely different combinations to test.

Use the full ninety characters where you have something to say. A thirty-character description wastes most of the space you were given.

TESTING PROPERLY, WITH A CHAMPION AND A CHALLENGER

Most accounts either never test ad copy, or test it so badly the results are meaningless.

The method that works on a small budget is a rolling champion and challenger.

How it runs

Your current best advert is the champion. Set A.

Write one challenger, Set B, and put it in the same ad group. Not three challengers. One, so the comparison is readable.

Let both run until each has accumulated enough impressions to mean something. Around a thousand each is a working minimum for a small account.

Then keep the winner, pause the loser, and write the next challenger. Set C against the surviving set. Then D. It’s a rolling ladder rather than a one-off experiment.

The gate that stops you from fooling yourself

If neither ad has reached a meaningful sample, the honest answer is inconclusive. Write that down and leave it running.

Declaring a winner on a hundred impressions is how accounts end up chasing noise for months, and it feels productive the entire time.

Keeping the sheet readable

If you manage this in a spreadsheet, group all of Set A together and all of Set B together, separated by a blank row and a repeated header.

Interleaving them makes the phased rollout impossible to read, and whoever picks the sheet up after you will not know which lines belong to which test.

And judge on qualified enquiries where you can, not click-through rate. A headline that promises the world gets clicked and wastes your afternoon.

WHAT TO CHECK BEFORE SEPTEMBER

A short list, given the migration window.

Find out whether AI Max is on, and whether your campaigns use campaign-level broad match or automatically created assets. Those are the ones migrating.
Check whether final URL expansion is enabled, because that is the setting that can bypass your pinned assets.
Read your landing page copy properly. It is now raw material for generated ad copy, so vague or outdated wording on the page can end up in an advert.
List anything that must always appear in your ads. Brand, regulatory wording, a required qualifier. Decide deliberately how you are protecting it.
Screenshot your current ad copy and your current performance. If something changes in September, you will want a before.
Check brand settings, which give inclusions and exclusions at campaign and ad group level.
Do not switch everything off in a panic. Some of this genuinely performs well, and the point is to choose rather than to be defaulted into it.

Half an hour now is worth considerably more than a confused fortnight in October.

A WORKED EXAMPLE, START TO FINISH

A hypothetical to show the shape. An independent letting agent, spending around a thousand pounds a month on Search.

Their adverts have eleven headlines. Nine of them are variations on “Trusted Letting Agents”, which is why ad strength says average and nobody has ever worked out why.

Nothing is pinned. Their company name appears in about a third of impressions and never in the same place twice.

They rebuild the copy properly.

Three keyword headlines drawn from their search terms report: “Letting Agents In Leeds”, “Property Management Leeds”, “Rent Your Property Out”.

Three benefit lines: “Tenants Found In 14 Days”, “Full Management From 8%”, “No Let, No Fee”.

Three objection lines: “No Long Tie-In Period”, “Fully Referenced Tenants”, “Rent Guaranteed Option”.

Three proof lines and three action lines complete the fifteen.

They pin by bucket. Keyword group to position one, benefits to position two, action to position three.

Ad strength drops from average to average, and they leave it because the structure is deliberate.

Then the AI Max check

Final URL expansion is on. They rely on the agency name appearing, so they make a considered decision about that combination rather than leaving it as a default.

They read their own landing pages and find one still advertising a fee structure they changed in the spring. That page is now feeding generated copy, so it gets corrected.

Then the test

Set B goes in a fortnight later with a different benefit angle, priced certainty rather than speed.

At roughly a thousand impressions each, Set B produces more enquiries at a similar click-through rate. B becomes champion, A is paused, C gets written.

Nothing clever happened. Somebody wrote fifteen genuinely different lines, kept the structure, and tested one thing at a time.

THE MISTAKES THAT WASTE THE MOST MONEY

Fifteen headlines that are the same sentence rephrased, which caps what any combination can achieve.
Pinning nothing, so the brand appears erratically and the advert reads differently every time.
Pinning everything, which leaves the system with almost no combinations to test.
Chasing an excellent ad strength rating by unpinning things you needed pinned.
Assuming pinning is a guarantee, when final URL expansion can bypass pinned assets.
Never reading your own landing pages, while they quietly feed generated ad copy.
Declaring a winner after a fortnight and a hundred impressions.
Running three challengers at once, so no comparison means anything.
Judging ad copy on click-through rate rather than on the enquiries it produced.

FREQUENTLY ASKED QUESTIONS

How many headlines does a responsive search ad need?

The minimum is three headlines and two descriptions.

You can supply up to fifteen headlines at thirty characters each, and four descriptions at ninety characters each.

Google assembles combinations and typically shows two or three headlines and one or two descriptions at a time.

Should I pin headlines in responsive search ads?

Yes, for anything that must always appear. Your brand, a regulatory phrase, a required qualifier.

Google advises against pinning identical or very similar text to the same slot, because it restricts the system.

Pin by group rather than by individual line, so each position always draws from a bucket and you keep the structure without killing the testing.

One caveat. With final URL expansion enabled, Google states that pinned assets are not used if a more relevant URL is chosen.

What is AI Max for Search campaigns?

A suite of targeting and creative features Google layers onto existing Search campaigns.

Its components are search term matching, text customisation, final URL expansion, locations of interest, brand settings, URL inclusions and exclusions, and reporting updates.

It is a layer on what you already run rather than a separate campaign type.

Does AI Max rewrite my ads?

Text customisation, previously called automatically created assets, uses text from your existing ads, your landing page copy and your assets, along with generative AI, to create ad copy matched to specific searches.

So yes, additional copy can appear that nobody at your business wrote.

That makes your website copy an input to your advertising, which is worth knowing before it happens rather than afterwards.

What is ad strength in Google Ads?

A rating from poor to excellent, based on the quantity and diversity of assets you supplied.

It’s a completeness measure, not a performance measure. It doesn’t know your conversion rate or your margins.

An advert rated excellent can easily sell less than one rated average.

How do I test responsive search ads properly?

One champion and one challenger in the same ad group. Not three challengers.

Let both reach a meaningful sample, around a thousand impressions each as a working minimum for a small account.

Keep the winner, pause the loser, write the next challenger. Repeat as a rolling ladder.

If neither reached a decent sample, the answer is inconclusive, and writing that down is more useful than inventing a winner.

Why did my ad text change without me doing anything?

Most likely text customisation, or an automatically created asset from before the naming changed.

Check whether AI Max is enabled on the campaign, and look at the asset reporting to see which assets are being served.

It’s a setting rather than a fault, and it can be reviewed.

WHERE TO START THIS WEEK

Two jobs, and the first takes ten minutes.

Open one of your Search campaigns and find out whether AI Max is on, and whether final URL expansion is enabled. Whatever you decide about them, decide it deliberately rather than inheriting it in September.

Then open your ad copy and count how many of your headlines are genuinely different from one another. If nine of the fifteen are the same promise in different words, that’s your afternoon.

Rewrite them in five groups. Keyword, specifics, objections, proof, actions. Three lines each.

The irony of all this is that the more Google writes, the more the words you supply decide what it has to work with. Automation doesn’t remove the writing job. It moves it upstream, into the assets and the pages the machine is reading.

Which means the businesses that still write carefully are about to have a bigger advantage than they did, not a smaller one.

Did You Enjoy This Blog Post?

I hope you enjoyed this blog post, and thank you so much for being here. We also upload videos to our YouTube channel every weekday. Please subscribe so you are one of the first to be notified.

If you enjoyed this blog, you may also like: