From Google Ads to ChatGPT Ads: Reusing What You Already Have

You already have hundreds of ad texts in Google Ads and Meta. The question is not whether they can be reused for ChatGPT Ads — they can — but how much of them actually carries over.

OpenAI has its own workflow for this, and it is more disciplined than you might expect. The core message is disarmingly direct: "You are not just rewriting Google or Meta ads." Here is the process, and where we think it is smarter than it looks.

The premise: build clean, launch at scale

OpenAI's stated minimum standard is at least 100 reviewed, upload-ready ads distributed across focused ad groups. That is a floor, not an aspiration.

But — and this is the important qualifier — they explicitly say scaling does not mean hundreds of tiny rewrites. It means covering meaningfully different customer intents, each tied to the right landing page and supported by real source material. A hundred near-identical ads hits the number and misses the point.

Step 1: Export what you already have

Start by exporting the ad tables from your existing platforms. OpenAI lists which fields are worth carrying over:

Platform	What to export
Google Ads	Active and recently high-performing search ads, including responsive search ad headlines and descriptions. Campaign, ad group, status, final URL, impressions, clicks, CTR, conversions.
Meta	Text-oriented and link ads. Exclude video-only creative unless the text can support future ChatGPT ad copy.
Microsoft Ads and others	Search, shopping text, native or sponsored content with reusable text and destination URLs.

Include paused ads only if they are still accurate and useful for brand voice. Exclude disapproved, expired or legally outdated ads — or label them clearly as examples not to reuse.

Step 2: Build an approved claims list

This step is what separates OpenAI's workflow from an ordinary "rewrite the ads" exercise, and it is the part we think is most worth copying regardless of channel.

Before any ad copy gets written, compile every claim, offer, price, ranking, certification, guarantee and comparison you intend to use — and evidence each one against either the exported ad or your website. The rule is simple: use no claim that is not supported by the source material.

The table should contain the claim, the supporting source, safe wording that can be reused, risk notes, and a clear yes / no / review required.

If the website and the old ads disagree, use the more current source or flag the item. If a claim is regulated, time-sensitive or unclear, mark it for review rather than using it freely.

The reason this step exists is that the workflow is designed to be run with an AI assistant. Without an approved list, the assistant invents plausible but unsupported claims. With it, the assistant is constrained to what you can actually stand behind.

Step 3: Build an intent map

The next step translates source material into how people actually ask for help in ChatGPT, rather than mirroring search keywords.

For each priority product, service or landing page you map: the user problem or need, intent cluster, audience or situation, constraint or tradeoff, and funnel stage — problem framing, discovery, comparison or purchase.

That map then becomes your ad groups. One row per meaningfully different buying situation, not per service in your price list.

Steps 4–6: Structure, context hints and ads

From here the workbook takes over. Campaigns group ad groups that share a business goal, budget, schedule and country targeting. Ad groups are built around one product area, theme, audience or intent area.

Context hints are written as short descriptive phrases of roughly four to five words, five to ten per ad group. We have written at length about how they work in our guide to building the campaign — including the mistake we made ourselves at first.

For the ads, titles run 16–24 characters and copy 32–48 characters. Build a clean launch set against your highest-value products first, then scale to a hundred in batches of no more than fifty. Prioritise the ad groups with the fewest ads so far and keep distribution as even as possible.

Step 7: Remove duplicates and score quality

This step is missing from most ad processes we see, and it is probably the most valuable part of the whole flow.

Before anyone reviews a hundred ads manually, every row is scored 1–5 on six dimensions:

• Source support — the claim is supported by ads or website
• Landing page fit — the ad's promise matches the destination
• Conversational intent fit — the copy maps to how users actually ask for help
• Distinctiveness — not a minor rewrite of another ad
• Clarity — clear value, audience, use case or next step
• Character compliance — title and copy within limits

Ads that are unsupported, misleading, too generic, repetitive, brand-first without user intent, mismatched to the landing page or outside the character limits get removed or flagged.

The point is to protect review time. It is wasteful to have a person read a hundred rows when thirty of them should never have reached them.

Steps 8–9: Validation and human review

Validation is mechanical: campaign names unique and populated, every ad group linked to an exactly matching campaign, context hints as valid JSON arrays, every ad row linked to an exact ad group name, characters within limits, URLs reachable, images public and square, no duplicate rows, no merged cells, and the workbook containing exactly the Campaigns, adgroups and ads tabs with no working tabs left behind.

One detail that is easy to miss and will block delivery: landing pages must not block OpenAI's user agents. Check your robots.txt before uploading. Ours is open, but it is worth verifying — particularly if you added AI crawler blocks in the past year, as many sites did.

And then the point OpenAI repeats hardest of all: passing formatting checks is not the same as being ready to launch. A qualified person on your team must approve every ad, context hint, landing page and image link for accuracy, completeness, policy compliance and brand fit before upload.

Measurement: a phase after launch, not a blocker before it

OpenAI is explicit that conversion tracking should be treated as a scale-phase next step — not as something that blocks a clean first launch. Launch with solid structure, enough distinct ads, validated context hints and approved landing pages. Add measurement once the campaign is live or close to it.

There are two paths:

	Conversions API	JavaScript pixel
Where it runs	Server to server	In the browser
Best when	You can send events from a backend, ecommerce platform, CRM or offline systems	You need a fast website-based setup or lack server-side access
Tradeoff	More reliable, but needs engineering support, event mapping and data handling review	Faster, but affected by browser limits, consent settings and script blocking

The recommendation is to start by defining the three to five conversion events that reflect real business value. Use the pixel for speed when server-side work is not immediately available. Use CAPI when reliability, backend events, offline or CRM events, or stronger attribution quality matter. If you run both, align event names and reuse event IDs so browser and server events can be deduplicated.

For Swedish companies the consent question is not theoretical. A client-side pixel blocked by consent banners or browser restrictions measures systematically wrong. If you already have server-side tracking in place, CAPI is the natural choice.

Why this step pays off

The material includes results from advertisers who moved to conversion optimization after starting with click campaigns:

• A software and technology advertiser shifted 73% of spend to Conversion Optimization: +22% CTR and 87% lower cost per order.
• Another in the same category shifted 98%: +27% CTR and 50% lower cost per order.
• A financial services advertiser shifted 70%: +4% CTR and 50% lower cost per lead.
• A retail and ecommerce advertiser put 40% of eligible spend into Conversion Optimization and saw 7% higher CTR and 65% lower cost per order than click-focused campaigns over the same period.

The figures are observed and may reflect other campaign changes — OpenAI says so in the footnotes themselves. But the direction is unambiguous and consistent across four independent advertisers in three categories. It is the strongest argument in the whole pack for not stopping at click optimization.

The most common mistakes among smaller advertisers

OpenAI lists them explicitly, and they are recognisable:

• Uploading only a few ads. Plan for at least a hundred before launch.
• Treating context hints like keywords. Write phrases describing real needs and situations.
• Unsupported claims. Use the approved claims list.
• Broken image links. Public, direct PNG or JPG links. Open every URL manually.
• Homepage-only landing pages. Map each ad to the most relevant product, service or content page.
• Template errors. Use header names, preserve tab structure, avoid merged cells.
• Skipping final review. Formatting checks are not approval.

Need help structuring the move from your existing channels? Book a free analysis and we will look at your accounts and what is worth carrying over.