What Angi’s ChatGPT Ads Results Prove—and What They Don’t

A new advertising platform does not become credible because one brand tries it.

It becomes more interesting when a sophisticated performance advertiser tests it, improves the economics, and expands.

That is what makes Angi’s ChatGPT Ads case study important.

Bottom line

Angi’s results are the strongest public evidence so far that homeowner-project intent inside ChatGPT can become measurable lead growth. They are not proof that an individual contractor will achieve the same CPL or ROAS. The lesson is to test narrowly, send better conversion signals, and judge the channel on booked work.

What Angi did

Angi began with a relatively simple test across roughly 20 home-service categories, including HVAC, roofing, and handyman.

As Ads Manager gained:

• daily budgets
• CPC bidding
• conversion optimization

Angi broadened the program into nearly 30 additional categories.

That expansion matters more than the existence of the first test.

Large performance teams run experiments constantly. Many do not scale.

The published results

OpenAI reports that Angi achieved:

• 38%+ improvement in ROAS quarter over quarter
• 34% lower cost per lead after launching conversion-optimized campaigns

Read OpenAI’s full Angi case study.

Angi’s vice president of search engine marketing described the value as reaching homeowners “not passively scrolling, but actively seeking answers.”

That phrase captures the strategic appeal better than a generic “high intent” claim.

A homeowner in ChatGPT may be:

• diagnosing a problem
• evaluating whether the project is urgent
• deciding between DIY and professional help
• comparing service approaches
• working out what type of contractor to call

That is project intent taking shape.

What the results prove

1. Home-services demand exists inside ChatGPT

This is no longer purely theoretical.

A company whose business depends on homeowner projects found enough qualified opportunity to expand across dozens of categories.

2. The channel can produce performance outcomes

The published metrics are not awareness metrics.

Cost per lead and return on ad spend are commercial outcomes.

That matters.

3. Measurement quality changes performance

Angi’s 34% CPL reduction followed the launch of conversion-optimized campaigns.

That supports a broader performance-marketing truth:

A platform can optimize more intelligently when it receives better signals about what the advertiser values.

Clicks tell the system who clicks.

Lead events tell it who becomes a lead.

Downstream data can create a clearer picture still.

4. Early platform changes are meaningful

Angi did not run one static campaign and wait.

Its strategy evolved as OpenAI released:

• more budget control
• CPC bidding
• conversion optimization

The current platform is already materially different from the earliest pilot.

What the results do not prove

This is where most commentary becomes sloppy.

They do not establish a contractor benchmark

Angi is a marketplace.

It has:

• national scale
• broad category coverage
• large data volume
• brand recognition
• a mature lead funnel
• enough conversions for algorithmic optimization

A 12-truck HVAC company serving one metro does not have those advantages.

They do not prove every home-service category works equally

Roof replacement, appliance repair, recurring cleaning, pest control, and handyman jobs have different:

• urgency
• job value
• lead volume
• conversion rates
• buying cycles
• margins

“Home services” is not one market.

They do not prove incrementality

Attributed ROAS is not the same as causal incremental ROAS.

A homeowner may see a ChatGPT ad and later arrive through branded search. Depending on measurement rules, the credit can move.

OpenAI is now working with incrementality partners and geo-based testing providers, which is a sign that the company recognizes this distinction. See the October measurement update.

They do not prove local inventory depth

A national advertiser can aggregate demand across many locations.

A local contractor needs enough eligible conversations in a defined service area.

That is why delivery itself is one of the first questions a local pilot must answer.

The local contractor actually has one advantage

Angi has scale.

The contractor has specificity.

A strong local HVAC advertiser can build the entire experience around:

• furnace replacement
• selected ZIP codes
• same-week estimate capacity
• financing
• local licensing
• the actual brands installed
• real warranty terms
• a local tracking number
• booked-job data in ServiceTitan

That can create a tighter connection between:

conversation → ad → landing page → appointment → sale

A marketplace has to serve many contractors.

The contractor only has to represent itself well.

What we would copy from Angi

Start with categories that matter

Do not spread money across every service.

Choose the service line with the clearest economics and capacity.

Move quickly when the platform adds capability

The companies that learn the channel early will develop operating knowledge before the market standardizes.

Angi’s own executive said early adoption helped the team learn quickly and build with the platform.

Optimize toward a business event

Once tracking is healthy and sufficient signal exists, move beyond traffic-only optimization.

The platform currently supports conversion optimization toward supported standard events. See OpenAI’s conversion-campaign documentation.

Expand only after evidence

Angi expanded after it saw improving economics.

That order matters.

Test.

Measure.

Then broaden.

What we would not copy

A local contractor should not launch 20 categories.

It should not imitate a marketplace landing experience.

It should not assume Angi’s published percentages.

And it should not define success as “being present in AI.”

The business case is sold jobs.

The real signal

The most important fact is not 38% or 34%.

It is this:

Angi treated ChatGPT Ads as a channel worth scaling, not a publicity experiment.

For a local home-service company, that is enough to justify a disciplined test.

It is not enough to skip the discipline.

For a complete overview of how ChatGPT Ads work for home services, and an honest breakdown of what a test actually costs, see our detailed guides.

Sources

• OpenAI Customer Story: Angi
• OpenAI Help: Conversion-Optimized Campaigns
• OpenAI Ads Blog: More Ways to Measure ChatGPT Ads