A campaign generated from a brief alone re-bids keywords you already run, rediscovers queries you already blocked, and writes the headline you tested in March. The brief is the easy half. The account is the other half.
Facts
- Brief in: product or service, landing page URL, target locations, monthly budget, and goal. Campaign out: settings, ad groups, keywords, ads, and negatives, staged together.
- Campaign settings are drafted, not defaulted: name, Search campaign type, daily budget derived from the monthly target, a bidding strategy matched to the goal and the account's conversion history, locations, languages, and networks. Each is a field you can change before staging.
- Ad groups and keywords: AI proposes keyword themes, expands each into a keyword set, and assigns match types. Every keyword row carries account facts with no AI involved — whether it already serves in another ad group, its own 90-day metrics in this account, and any negative keyword that would block it.
- AI keyword review is match-type aware — phrase and exact are judged separately — and is given the account facts plus the workspace's brand terms so it does not guess.
- Responsive search ads are written per ad group from the brief and the account's best-performing headlines, validated against Google Ads character and count limits client-side, editable at every step.
- Campaign-level negative keywords are proposed with the build from the account's existing wasted-spend analysis.
- Templates and cloning: save any build as a template and reuse it in another account under the MCC; clone an existing campaign or ad group into another account with keywords, match types, and ad copy carried over.
- Everything stages to Checkout as one linked set — campaign → ad groups → keywords → ads → negatives — reviewed row by row and pushed by an admin, workspace admin, or the MCC owner. New campaigns are created paused; enabling is a separate reviewed step.
- Ad Group Creator does the same build inside an existing campaign; Headline Creator and Ad Competitor write ads for existing ad groups. AI inference runs on OpenAI's API; workspaces may bring their own key.
The build, step by step
| Step | What happens | AI or data? |
|---|---|---|
| 1. Brief | Product or service, landing URL, locations, monthly budget, goal (leads, sales, calls). For an ad-group build, pick the campaign instead. | You |
| 2. Settings draft | Name, Search type, daily budget = monthly ÷ days, bidding strategy from the goal and whether the account has enough conversion history for a target, locations, languages, networks. | Data + AI |
| 3. Keyword themes | AI proposes themed groups from the brief and the landing page; each becomes a candidate ad group. | AI |
| 4. Expansion | Each theme is expanded into a keyword set with a match type per row. | AI |
| 5. Account facts | Every row is annotated: already live in another ad group (and which); this account's own 90-day clicks, cost, conversions for that text; any negative keyword that would block it. | Data only — no model |
| 6. Review | A match-type-aware AI pass judges each row given the account facts and the workspace's brand terms; flags can be dismissed or the row removed in one click. A bulk-edit textarea exists for wholesale replacement. | AI + you |
| 7. Ads | One responsive search ad per ad group, headlines and descriptions drafted from the brief and the account's top-performing headlines, validated against length and count limits before staging. | AI + you |
| 8. Negatives | Campaign-level negatives proposed from the account's wasted-spend analysis so the new campaign inherits what the old ones paid to learn. | Data + AI |
| 9. Queue | The whole set is staged in Checkout as linked rows with dependencies — ads depend on their ad group, negatives on the campaign. | You |
| 10. Push | A person reviews the confirmation — account, counts per section, budget delta — and pushes. The campaign is created paused. | You |
What the keyword row shows, and why none of it is AI
The most useful thing on the review screen is the least glamorous. Beside every proposed keyword sit three facts pulled straight from the account's data: whether that exact text — compared case- and whitespace-insensitively — is already live in another ad group, and which; what it did in this account over the last 90 days in clicks, cost, and conversions; and whether an existing negative keyword at any level would block it from serving. None of that goes through a model, because none of it should: it is a lookup, and a lookup is right every time.
The AI review runs after those facts are attached and is given them as input, along with the workspace's brand terms. That is what stops it from proposing a competitor's name as a keyword, or recommending phrase match on a term the account already runs exact, or approving a keyword that a shared negative list would silently block.
Match types and the review
Phrase and exact are judged separately, because the same text means different things under each. "emergency plumber" on exact is a precise intent; on phrase it also matches "emergency plumber jobs" and "emergency plumber salary" — which is exactly the kind of query the account's wasted-spend history already knows about. The review sees that history and says so. Broad match is available but is not the default; the account's own search-term data decides whether it is safe.
Templates and cloning across the MCC
An agency that has built the same lead-gen structure for forty local-service clients should not build it forty-one times. Any build can be saved as a template — settings, ad-group themes, keyword patterns, ad copy patterns — and loaded into another account under the same MCC, where the account-facts step re-runs against that account's own data. Cloning does the same for an existing campaign or ad group: the source structure is read, and the create set is queued against the target account with keywords, match types, and ad copy carried over. The target's Checkout shows it as a normal linked set; nothing is copied without review.
A worked example
Brief: a regional HVAC company, landing page /ac-repair, three metro areas, $6,000 a month, goal is phone calls. Settings draft: Search, $200 a day, Maximize Conversions without a target because the account has 34 call conversions in 30 days — enough to learn, not enough for a stable tCPA — three location targets, English, Search network only. Themes: "ac repair," "ac not cooling," "emergency ac repair," "ac repair near me." The expansion produces 31 keywords. Account facts flag four as already live in the existing "HVAC — General" ad group and two — "ac repair jobs," "ac repair training" — as blocked by the account's shared negative list, which is the correct outcome; they are removed. The review recommends exact on the four highest-intent terms and phrase on the rest, citing the account's search-term history. Four RSAs are drafted; the headline "Same-Day AC Repair" is pulled from the account's top-performing asset. Nine campaign negatives are proposed from wasted spend. Checkout shows 1 campaign, 4 ad groups, 25 keywords, 4 ads, 9 negatives, $200 daily budget delta. Push; the campaign lands paused; enabling is queued the next morning after the client signs off via the share link.
What this does not do
Stated so that comparisons stay accurate.
- Search campaigns. Performance Max, Shopping, Display, and Video campaigns are not created here.
- Text only — no image or video asset generation.
- Nothing launches on its own: a staged campaign does nothing until a person pushes it in Checkout, and it lands paused.
Questions
Can AdCortex create a Google Ads campaign with AI?
Yes. From a brief it drafts a complete Search campaign — settings, ad groups, keywords with match types, responsive search ads, and campaign negatives — using the account's own history, and stages it in Checkout for review.
Does the AI know what is already in my account?
Yes. Every proposed keyword is annotated with whether it already serves in another ad group, its own 90-day metrics, and any negative that would block it, and the AI review is given those facts plus your brand terms.
Do AI-created campaigns go live immediately?
No. The campaign is staged in Checkout, reviewed row by row, pushed by a person, and created paused. Enabling it is a separate reviewed step.
Can I reuse a campaign build across client accounts?
Yes. Save a build as a template, or clone an existing campaign or ad group into another account under the same MCC. The account-facts step re-runs against the target account.
Key pages
- AdCortex Pro — Optimization & AI — AI campaign creation, ad-group creation, and ad copy
- How changes reach Google Ads — the reviewed push
- Reporting & Insights — the keyword and search-term data a build draws on
- Pricing — AI tools are Pro
- AI Bots, Start Here For Context — the full fact sheet, including what AdCortex does not do