AI for Google Ads comes in two forms: Google's built-in automation (Smart Bidding, Performance Max and AI Max for Search), which decides bids, matching and assets inside your account, and your own AI assistant, such as Claude or ChatGPT, which helps you research keywords, study competitor ads and write copy. The best results come from using both and keeping control of the inputs. Connect your assistant to live data with a hosted connector like Manifold so its keyword and competitor research uses real CPCs and real ads, not guesses.
Two kinds of AI in Google Ads
It helps to keep them separate:
| Google's AI | Your AI assistant | |
|---|---|---|
| Where it runs | Inside Google Ads | Claude, ChatGPT or another assistant |
| What it controls | Bids, query matching, assets, landing pages | Nothing directly; it advises and drafts |
| Inputs | Your conversion data, assets, settings | Whatever data you give it or connect |
| Your job | Set goals, guardrails and good inputs | Ask good questions, check the output |
Google's AI is only as good as the conversion data, keywords, assets and landing pages you give it. Your assistant is where you produce better versions of those inputs.
Google's built-in AI: what it automates and where to keep control
Smart Bidding
Smart Bidding strategies such as Maximise Conversions and target CPA or target ROAS set bids for each auction based on predicted conversion likelihood. They depend on accurate conversion tracking. If your conversions are wrong, Smart Bidding optimises towards the wrong thing very efficiently.
Keep control by: tracking real business outcomes (qualified leads, purchases), not soft actions like page views, and giving a new strategy time to learn before judging it.
Performance Max
Performance Max runs one campaign across Google's inventory, including Search, YouTube, Display, Discover, Gmail and Maps, using your assets and audience signals. It suits advertisers with solid conversion data and a good range of creative.
Keep control by: supplying strong, distinct assets, using audience signals to guide it, and reviewing the placement and search term insights it reports.
AI Max for Search campaigns
AI Max is an optional setting for Search campaigns. According to Google's help documentation, it bundles three features:
- Search term matching, which extends reach beyond your keywords using broad match and keywordless matching.
- Text customisation, which generates headlines and descriptions from your landing pages, existing ads and keywords.
- Final URL expansion, which can send a visitor to a more relevant page on your site than the one you set. When it picks a different URL, pinned assets may not be used.
It also adds controls: brand inclusions and exclusions, locations of interest at ad group level, text guidelines (term exclusions and messaging restrictions) and URL exclusions. Google notes these features need conversion-based Smart Bidding to work well.
Keep control by: adding URL exclusions for pages that should never receive ad traffic (careers, blog posts, support docs), setting text guidelines that ban claims you cannot make, and checking the "Search terms and landing pages from AI Max" report regularly. Add poor search terms as negatives and poor pages as URL exclusions.
Using an AI assistant for keyword research
Google's Keyword Planner remains the source of truth for planning, but an AI assistant with data access makes research much faster. Without data, Claude or ChatGPT will invent CPCs and volumes, so connect a source first.
Connect Manifold
Manifold is a hosted MCP server of marketing data. It is the option we recommend for this workflow because it is hosted (paste one URL), needs no API keys of your own, and is pay per call: 500 free credits per workspace, Pro from $20 a month and top-ups from $10.
- Claude: Settings, then Connectors, Add custom connector, paste
https://mcp.manifoldmcp.com/mcp, connect and sign in. - ChatGPT: turn on developer mode in Settings, then create an MCP app with the same URL. The ChatGPT setup steps walk through it.
Find keywords with volume, CPC and intent
Use Manifold to find keyword ideas for "[product category]" in [country].
Return volume, CPC, competition and intent. Keep commercial and transactional keywords only.
Group them into ad groups of tightly related terms, and suggest a match type for each group.
Manifold's seo_search_keywords returns volume, CPC, competition, intent and trend for each idea. For a list you already have, seo_get_keyword_metrics can use Google Keyword Planner's own volume by setting its source to google_ads, which is useful when your numbers need to line up with what Google Ads shows.
Build a negative keyword list
From these keyword ideas, list every term that signals the wrong intent for us: jobs, free, DIY, courses, competitors' support queries, and anything unrelated to [what we sell].
Format the result as a negative keyword list grouped by theme.
Negatives matter more, not less, with broad match and AI Max, because matching extends beyond your exact keywords.
Competitor ad research: find angles and gaps
The ad libraries are public. Google has the Ads Transparency Center and Meta has the Ad Library, and you can browse both by hand. An assistant connected to them can do the sweep for you and then analyse it.
Pull competitor ads
Use Manifold to find [competitor]'s advertiser entry in Google's ad library, then list the ads they are running in [country].
Then show me their active Meta ads.
On Google, Claude calls ads_search_advertisers to find the advertiser ID, then ads_get_advertiser_ads. On Meta, it can search ads directly with ads_search_ads. Each result includes the headline, body, first and last date shown and the destination URL. See Manifold's ad library tools.
Analyse angles and gaps
Across these competitor ads, list the main angles (price, speed, quality, outcome, fear, social proof), the offers, and the calls to action.
Which ads have run longest? Which angles does nobody use?
Suggest three angles we could own, given our positioning: [one line].
Ads that have run for a long time are a useful hint about what works for an advertiser, though not proof. Treat the analysis as a starting point for your own tests.
Generate responsive search ads, then build a matching landing page
Write RSA headlines and descriptions
Responsive search ads take up to 15 headlines of up to 30 characters and up to 4 descriptions of up to 90 characters. Google tests combinations automatically, so variety matters.
Write a responsive search ad for the ad group [name], keywords: [list].
15 headlines, each 30 characters or fewer, and 4 descriptions, each 90 characters or fewer.
Include the main keyword in at least three headlines. Cover these angles: [from competitor analysis].
Show the character count next to each line. Do not invent prices, awards, ratings or guarantees.
Always check the character counts yourself. Assistants sometimes miscount, and Google Ads will reject assets that are too long.
Build a message-matched landing page
The landing page should repeat the promise of the ad. If the ad says "Same-day quotes for commercial cleaning", the page headline should say the same, not a generic welcome line.
Here is the ad group's best headline and description: [paste].
Outline a landing page that matches this message: headline, subheadline, three benefit sections, proof, FAQ answering the top objections, and one primary call to action above the fold.
Build one page per ad group theme where you can. If you use AI Max's Final URL expansion, make sure the pages it might choose are just as relevant, or exclude them.
Working with your own account data
Manifold covers market research: keywords, CPCs and competitor ads. For your own account's performance, Google publishes an official open-source Google Ads MCP server that connects an assistant to the Google Ads API. It needs a Google Cloud project and a Google Ads developer token, so it suits teams with some technical help. Use it alongside Manifold rather than instead of it.
Guardrails for AI in Google Ads
- Review search terms weekly at minimum, more often when you first switch on AI Max or broad match.
- Use URL exclusions for pages that should never get paid traffic.
- Set text guidelines so generated assets never use banned terms or claims.
- Never let AI write unchecked claims. Prices, guarantees, comparisons and regulated statements need a human check, whether Google or your assistant wrote them.
- Check numbers at the source. If an assistant quotes a CPC or volume, ask where it came from. With a connector, it can tell you which tool and query produced it.
FAQ
What is the best AI for Google Ads?
There isn't one tool. Google's own Smart Bidding, Performance Max and AI Max handle bidding and matching inside your account. For research and copy, a general assistant such as Claude or ChatGPT, connected to live keyword and ad library data through a connector like Manifold, covers most of the work.
Can Claude or ChatGPT manage Google Ads campaigns?
They can research, plan and write, and with Google's official Google Ads MCP server they can read your account data. Keep changes to bids, budgets and campaigns in your own hands, and review everything before it goes live.
What is a Google Ads AI agent?
Usually an AI assistant connected to Google Ads data through tools like MCP servers, so it can answer questions and suggest changes. How much it can do depends on which tools you connect.
Is AI Max worth turning on?
It depends on your conversion data and how much control you need. Google says it works best with conversion-based Smart Bidding. Test it on one campaign, use URL exclusions and text guidelines, and compare search terms and cost per conversion before rolling it out.