Skip to main content

How AI generates ad copy

When you reach the ad copy step of the campaign wizard, the Ad Copy Agent drafts the responsive search ad for your campaign, 15 headlines and 4 descriptions, from your product context, the keywords in the cluster you picked, and the intent type you chose. You review and edit every line before launch. The single biggest lever on the result is how specifically your product description (and the product context derived from it) describes what your product does.

The inputs​

The Ad Copy Agent combines four things to write the ad:

  1. Your product context. What you sell, who buys it, the jobs they hire it for, and the category it's in, derived from your product's website during onboarding.
  2. The keywords from the cluster you picked. What customers are actually searching for.
  3. The campaign's intent type. Non-branded, branded, or competitor, chosen on the targeting step. Each type uses a different copy playbook (see below).
  4. The destination URL and any extensions you've attached. Where the ad sends traffic and what auxiliary text shows alongside it.

On top of those, the agent works within Google's structural constraints for a responsive search ad:

  • Headlines: up to 30 characters each (the agent targets 24–30).
  • Descriptions: up to 90 characters each (the agent targets 80–90).
  • A character allowance for the display-path fragments shown next to your domain.
  • Google's content policies, no excessive caps, no unverifiable claims, no trademark misuse.

What it produces​

The Ad Copy Agent produces one responsive search ad: 15 headlines and 4 descriptions, written so that:

  • They address the search intent of the cluster's keywords. An ad for "alternative to jira" speaks to switching, not first-time discovery.
  • They reflect your product's specific positioning. What's distinctive, what jobs it does, and which customer it's for.
  • They give Google room to mix and match. Google rotates through the 15 headlines and 4 descriptions and assembles the live ad in real time, picking the combination it predicts will perform best for each query, device, and audience.

You don't get several finished ads to choose between, you get one ad's worth of headline and description options that Google recombines on the fly. Supplying many headlines rather than one fixed ad is what lets Google's algorithm optimize.

Why the intent type changes the copy​

The same product, the same keywords, and the same product context produce different ad copy depending on the campaign's intent type, because the searcher's intent differs. A branded search ("linear pricing") wants confirmation; a competitor search ("jira alternative") wants a credible alternative; a non-branded search ("project management for engineers") wants a fitting solution. Copy that's right for one is wrong for the others. There are three intent types, each with its own playbook:

  • Branded. Lead with the brand name and reinforce trust and officialness. CTAs are brand-anchored ("Try Linear Free", "Linear Pricing"). The searcher is already considering you, the ad's job is to confirm and close, not to introduce.
  • Competitor. Lead with comparative angles, but never name the competitor in headlines or descriptions. CTAs lean on reducing switching cost ("See How We Compare", "Free Migration"). The searcher is shopping, the ad's job is to position you as the credible alternative.
  • Non-branded. Lead with category language and value props. CTAs are generic ("Try Free", "Get a Demo", "See Pricing"). The searcher is solving a problem, the ad's job is to present you as a fitting solution and earn the click.

The biggest practical difference is in competitor campaigns, covered next.

Competitor campaigns and Google's trademark policy​

For a competitor campaign, Hero Marketer never puts the competitor's name in any headline or description. Google's policy allows a trademark in ad text only if you're the trademark owner, an authorized reseller, or an informational site about the trademarked product, none of which applies to a competitive campaign. Comparative phrases like "[Competitor] Alternative" or "Switch from [Competitor]" are common and often pass automated approval, but the trademark owner can file a complaint with Google. When that happens:

  • The ads get disapproved.
  • The restriction applies on an ongoing basis to any ad on the same domain.
  • You generally can't appeal without permission from the trademark owner.

The keyword side is different. Bidding on a competitor's brand as a keyword is allowed, and that's how your ad shows up on those queries in the first place. So the wizard targets the competitor's terms as bid keywords and writes copy that positions you as the alternative without naming the competitor in the visible text. A typical competitor ad might include headlines like:

  • "Top SaaS Alternative" (instead of "[Competitor] Alternative")
  • "Better Built. Less Cost." (a differentiator)
  • "Switch in Minutes" (a switching CTA)
  • "Free Migration" (an offer)

The competitor's name appears in your bid keywords, where it's allowed, but not in your ad text, where it isn't, and your ads still serve on competitor queries. If you have written authorization from a competitor to use their trademark in ad text, you can edit the headlines manually after generation. That's an unusual situation and generally doesn't apply.

Why product description quality matters most​

The single biggest lever on copy quality is your product description, because the agent can only write specifically about your product to the extent your description says specifically what your product does.

A thin description:

Linear is a project management tool. Easy to use and fast.

produces generic copy that could apply to any project management tool, headlines like "Powerful project management" or "Manage your work better."

A rich description:

Linear is a project management tool built for software teams. It replaces tools like Jira and Asana for engineers, designers, and product managers at series A through D startups, typically 20 to 200 people. The job teams hire Linear for is shipping software faster without bureaucracy. Customers describe their previous tools as slow, cluttered, and full of fields no one fills in. Linear's draw is speed, opinionated workflows, and a keyboard-first interface that makes triage and status updates near friction-free.

produces copy that names actual jobs ("Ship software faster"), names competitor pain ("Faster than Jira"), and reflects positioning ("Keyboard-first, built for engineering"). A richer description doesn't just produce more accurate copy, it produces copy that converts, because it speaks to what real customers care about. For what makes a good source page and what gets extracted from it, see Add your website and What product context is.

Why the agent generates keywords first, then copy​

The order matters: keywords define intent, and copy is written to match that intent. If you wrote copy first and chose keywords later, you'd end up with copy that doesn't speak to the queries triggering it. Hero Marketer goes keywords → copy because that's the order Google's auction works in, someone types a query, then Google picks an ad.

What the agent is weak at​

A few things the Ad Copy Agent doesn't do well, where you'll want to edit:

  • Distinctive brand voice. A very specific voice (Wendy's Twitter, Innocent Drinks tone) isn't captured by the agent's defaults. Edit headlines to inject brand voice where it matters.
  • In-jokes and inside references. The agent writes general best-practice prose. If your audience expects a specific cultural reference, add it yourself.
  • Aggressive claims. The agent defaults to defensible language. If your strategy is bold, policy-compliant hyperbole, you'll likely rewrite.
  • Compliance-specific copy. Legal disclaimers and regulated-category language need a human review.

For all of these, edit. The wizard supports inline edits to every headline and description, treat the output as a strong first draft, not a final.

Regenerating the copy​

If the first set isn't right, regenerate it for a new set with different angles. Before you do, decide whether the problem is the input or the output: if the description or keywords are off, fix those first and the copy will follow; if the description and keywords are right but the angles aren't compelling, regeneration gives you fresh angles. A campaign with weak inputs produces weak copy no matter how many times you regenerate. Credits are charged during the AI generation steps, so each regeneration spends credits, see How credits work.

Next​