On AI for ad operations

AI Max removes keyword targeting and lets Gemini match intent to your landing pages. It optimises brilliantly inside the budget you give it — and has no view of whether that budget belongs there.

Advantage+ automates audience, placement and budget end to end, and now generates creative variations too. What it can't do is tell you Meta isn't where the next dollar belongs.

TikTok now ships both an in-platform automation suite and its own MCP server for external agents. Here's what each covers and where a harness still adds something neither does.

Sidekick is a genuinely capable agent inside the Shopify admin — theme edits, segments, Flow workflows, proactive suggestions. It also cannot see a cent of your ad spend.

GA4 surfaces anomalies and generated insights on its own schedule, about its own data. The difference is who chooses the question — and whether the answer can include anything outside GA4.

Auto ads, Auto optimize and Smart Pricing are genuinely good machine learning — at placing and pricing ads inside AdSense. Here's the work they structurally don't do, and why that gap isn't a flaw.

A practical workflow — diagnosing an earnings drop by segment, ranking sites and ad units by RPM, catching policy issues before they cost money, and reconciling payments.

Google optimized floors, bidding eCPM floors and open bidding do a real job inside AdMob's auction. What they don't do is tell you which ad source is quietly failing to fill, or which app is carrying the others.

A workflow for reading the AdMob mediation report properly — observed eCPM per source, fill-rate gaps, the iOS/Android split, and which apps are carrying the portfolio.

Target CPM, Dynamic Allocation, First Look and Optimized Competition decide which demand wins each impression. None of them tell you which inventory, site or child publisher is worth your attention.

A workflow for reading a Google Ad Manager network — ranking ad units by yield, catching site problems, tracking MCM child publishers, and reviewing what's actually being served.

Google's official Google Ads MCP server is free, open-source and strictly read-only. Here's exactly what that buys you, what it structurally cannot do, and where an agent harness earns its place.

A GA4 MCP server answers GA4 questions well. The problem is that most questions worth asking about GA4 data need Google Ads, Search Console or your revenue data in the same breath.

Unlike Google's read-only server, Meta's MCP can change your account — and its paused-creation safety net is a genuinely good design. Here's where a single safety net stops being enough.

SHOPLINE (International) connects natively — store info, products, orders, an exact sales summary and anomaly detection, with store changes behind the same approval gate as every ad platform.

A concrete workflow for a SHOPLINE merchant — weekly sales review, product-level winners, anomaly alerts, and joining store revenue to ad spend for a profit number you can act on.

Ads Data Hub is Google's privacy-safe environment for querying event-level ad data you can't export. What it does, the aggregation rules that shape every query, and the far more common case where BigQuery is the right answer instead.

Which AdMob scope to request, how Google's scope tiers affect whether you need verification, and the practical lesson we learned choosing narrow scopes across twelve Google integrations.

First user CM360 advertiser name, CM360 placement name, campaign ID and the rest — what GA4's Campaign Manager 360 dimensions mean, how the two scopes differ, and the usual reason they come back empty.

Live per-token pricing for the lowest-cost frontier-adjacent models, how DeepSeek, Qwen, Gemini Flash and GPT-5.6 Luna compare, and the three jobs where paying more actually returns the money.

A practical guide to putting an AI agent on real SEO work — content-gap discovery, ranking-drop monitoring, cannibalization checks — and an honest account of the tasks where agents still make things worse.

They all sit at your domain root and they are constantly confused. What each file actually does, which crawlers honour which, and whether llms.txt is worth adding.

Platform ROAS ignores margin, returns and shipping — so the campaign it calls your best is often not. What data you need to optimize on net profit, and how to get it into the loop.

Looker Studio is free, flexible and slow to maintain. Where an agent-built dashboard is a genuine replacement, where it isn't, and the honest trade-offs between building a report and describing one.

What belongs on a Google Ads dashboard, how to get one built in minutes instead of an afternoon, and how to put it on a schedule so nobody has to remember to update it.

Monthly client reporting scales badly — the same report rebuilt per client, per month, by hand. How to make it a standing page per client instead, and what to check before sending a link outside your company.

Email an AI agent a task, get back a summary of what it would do, reply "confirm", and read the report in the same thread. A walkthrough of the Agent Mail workflow and where it beats opening the app.

An email address is trivially spoofable, and an inbox is full of text written by strangers. What a safe email-to-agent design has to defend against, the checks to demand before you connect one, and how Agent Mail answers each.

Inbox filters, no-code automation like Zapier or Make, and an agent mail inbox you can email directly are three different answers to "handle this email for me". What each is actually good at, and where each breaks.

Three tools for automating Meta and social ad buying, compared on their own terms — creative analytics, rule-based automation, and enterprise creative production — plus where each one actually fits.

A concrete walkthrough of what a Google Ads agent does for an online store — Merchant Center feed health, Performance Max asset groups by product category, and joining ad cost with your own Shopify revenue for true ROAS.

For service businesses and B2B, the whole campaign lives or dies on whether a form fill is actually counted as a conversion. A walkthrough of how an AI agent audits lead-gen conversion tracking, Quality Score, and cost per lead together.

How an AI agent monitors auction insights for competitors bidding on your brand terms, builds a defensive branded-search campaign, and tells the difference between a real threat and normal auction noise.

High-competition seasonal windows break the normal rhythm of budget pacing and bidding. How an AI agent checks pacing more frequently, defends against CPC spikes, and reverts cleanly once the spike ends.

Starting a Google Ads account from zero is a different job than optimizing an existing one — no historical data to diagnose, no search-terms report to mine. The exact sequence an AI agent follows to launch a new account correctly.

A step-by-step guide to connecting your Composio API key to Agent Planners — where to get the key, how it's stored, how to refresh the catalog after connecting a new app, and how to disconnect it.

Real, live per-token pricing across seven current models — from DeepSeek V4 Flash at $0.20/1M to Claude Fable 5 at $10/1M input — and what the 50x spread actually means for choosing a model per agent.

BYOM zero-rates LLM tokens but still charges a 15% platform-infrastructure fee on the same token volume. Here's the actual math on when that trade beats metered credits — and when it doesn't.

A concrete checklist for evaluating an AI ad-ops or benchmarking tool before you commit — write governance, first-party data joins, native platform reach, and how it actually integrates with GA4 and the rest of your stack.

Three well-known Google Ads optimization tools, compared on their own terms — rule engine depth, guided recommendations, and automated auditing — plus where an AI agent that executes end-to-end fits alongside them.

A practical walkthrough of what an AI agent connected to Search Console can actually do — finding content-gap opportunities, tracking ranking drops, and catching keyword cannibalization against your paid spend.

Agent Planners ships 24 built-in Google Ads skills — audits, Performance Max grading, Quality Score diagnosis, remarketing architecture and more. What they are, the standout ones, and how the agent decides which to use.

Agent Planners' Meta Ads skill is a single, comprehensive reference covering the whole Marketing API — entity reads, insights recipes, create/update/pause patterns and the errors to avoid. Here's what's in it and how it's used.

Agent Planners' two DV360 skills split cleanly along DV360's own structure: campaign operations (the create/update entity hierarchy) and reporting & reach planning. What each one teaches the agent, and how they combine.

TikTok Ads API Patterns is Agent Planners' single reference skill for the TikTok Business/Marketing API — reads, integrated-report recipes, mutation patterns and major-unit budgets. What's in it and how it's used.

Shopify Store Analysis is Agent Planners' built-in skill for reading store data correctly across both connection tiers — sales trends, product performance, customer cohorts — and connecting it to ad-platform spend.

CM360 Reporting & Analysis is Agent Planners' built-in skill for the ad-server side of Google Marketing Platform — the user-profile concept, exact v5 reporting vocabulary, entity audits, and the narrow five-collection trafficking write surface.

Managing dozens of Google Ads sub-accounts under one manager account (MCC) means repeating the same audit or optimization by hand, per account. Here's how an AI agent fans one goal out across a whole MCC, with a dry-run preview per account.

Before a DV360 line item ever goes live, reach forecasting answers the budget-vs-audience question. Here's how an AI agent runs that workflow end to end — plannable products, forecast simulation, targeting lookup.

CM360's role is cross-channel ad-server truth — but pulling and reconciling its reports by hand doesn't scale. Here's the concrete workflow for automating CM360 reporting and entity audits with an AI agent.

DeepSeek V4 Flash pairs a 1-million-token context window with some of the lowest per-token pricing on the AI Gateway. Real current pricing, where it fits among Agent Planners' six agents, and how new models like it surface automatically.

Anthropic's newest models, Claude Fable 5 and Claude Opus 5, are live in Agent Planners' model picker — with real credit pricing, a 1M-token context window, and no waiting for us to add them.

New Agent Planners organizations now default to GPT-5.6 Luna on every quality-critical agent — 13x cheaper than the model it replaces, on a bigger context window. Here's the pricing story, including the billing bug we found and fixed the same day.

Six agents, 40+ selectable models, one AI Gateway. A concrete framework for which model tier to run on the orchestrator versus a specialist — with real current pricing for GPT-5.6 Luna, Claude Fable 5, Claude Opus 5 and more.

Most AI ad tools built for Meta are creative-analytics specialists or rule engines. Here's what to look for in an AI agent that reads, reasons about and executes changes across your Facebook and Instagram campaigns.

TikTok's campaign → ad group → ad hierarchy and creative-first culture need different judgment than Search or Meta. What to look for in an AI agent for TikTok Ads, and where Agent Planners fits.

SEO gets you ranked; GEO gets you cited. The concrete, checkable techniques — structured data, llms.txt, answer-first writing, freshness — for showing up in ChatGPT, Perplexity and AI Overviews.

llms.txt is a plain-text file that gives AI systems a concise index of your site. What it actually is, what to put in it, and a real, live implementation you can inspect right now.

Ten concrete questions to ask before connecting any AI tool to a live ad account — approval gates, credential storage, magnitude limits, audit logs and more. Vendor-neutral, usable for evaluating any tool, including ours.

Eight criteria that separate a real Google Ads AI agent from a dashboard with a chat box — reviewability, write governance, whose numbers it trusts — plus the red flags and where each category of tool fits.

We scored 10 AI ad-ops tools — ourselves included — on write governance, first-party revenue joins and native platform reach, using each vendor's own public claims. Only one documents mandatory human approval on every write. Full dataset included.

A practical, week-by-week plan for rolling out an AI agent on a live ad account — connecting platforms, starting read-only, setting guardrails, turning on approval-gated writes, and scaling to a full account or MCC.

Ads Data Hub gives you privacy-safe access to Google's log-level ad-exposure data; BigQuery is a general-purpose warehouse for your own data plus whatever platforms export to it. Most ad-ops teams need BigQuery — here's how to tell.

RPA scripts click through the UI exactly as instructed; AI agents reason about a goal through the API. Here's the real distinction for Google Ads, and why RPA is the riskier choice on a live ad account.

Composio gives an AI agent 250+ pre-built SaaS actions with one API key. A custom MCP server exposes your own tools and data on an endpoint you host. Here's how to decide — and how to connect either one.

Most "AI SEO" tools generate text. The useful ones read your Search Console data, find the gap between what you rank for and what you sell, and keep a human between the idea and the publish button.

GA4 made reporting harder, not easier. Here's how to judge an AI agent for Google Analytics — whether it queries the API directly, handles GA4's data model honestly, and can join analytics to cost and revenue.

Most AI ad tools are built for Search and social and don't touch DV360 at all. Here's what a Display & Video 360 agent has to understand — the IO/line-item hierarchy, Bid Manager reporting, and why new line items should start as drafts.

CM360 is the ad server, not a buying platform — which changes what an AI agent should do with it. Trafficking, Floodlight, the asynchronous reporting lifecycle, and why writes here need the tightest gate in your stack.

Land third-party data in BigQuery, model it so an AI agent can reason over it, then capture your analysis method as a reusable Skill so every future run follows it instead of improvising.

Shopify knows what sold, Search Console knows what people searched, Google Ads knows what you paid. Wire all three to one AI agent and run a ten-minute daily loop — with every store change approved by you.

An MCP server hands an AI agent your Meta Marketing API — with no approval step, no team roles, no audit log and no spend ceiling. What that actually costs, and what a governance layer adds on top.

Agent Planners now exposes a REST API and an MCP server, so Claude Desktop, Claude Code, Codex, or any agent or script you run can create and monitor tasks against your ad accounts — with the same approval gates and guardrails as the in-app Workspace.

Most accounts quietly leak budget on zero-converting search terms, low-ROAS campaigns and dead placements. How an AI agent finds the waste and proposes the cuts — with your approval.

A complete Google Ads audit checklist — structure, search terms, budgets, tracking, disapprovals, wasted spend — and how an AI agent runs it in minutes, read-only.

What a multi-agent system is, why an orchestrator plus specialists beats one big prompt, and where it fits in marketing — from research to optimization to reporting.

Frontier models are powerful but pricey; cheaper models handle most ad-ops work fine. How to match the model to the task and keep your credit spend low.

Turn one-off AI tasks into scheduled jobs — weekly reviews, disapproval monitoring, budget pacing, search-term sweeps — that re-evaluate live data and email you what to approve.

Agentic AI, MCP, reviewability and first-party ROAS — the five shifts that actually changed paid search in 2026, and what they mean for your ad ops.

Rules and scripts do exactly what you wrote; agentic AI reasons from a goal. Here's the real difference for advertisers — and when each one wins.

Hand an AI your ad budget with no gate and it will eventually make an expensive, confident mistake. The case for human-in-the-loop — without killing velocity.

A practical playbook for using AI to remove the repetitive Google Ads work — audits, negatives, budget pacing, reporting — without losing control of the account.

How agencies use AI to run audits and optimizations across many client accounts (MCC) — with per-account guardrails, approvals and a dry-run so nothing goes wrong.

Performance Max automates reach but hides its reasoning. Here's how to keep visibility and control — with reviewable AI, first-party ROAS and negative-list discipline.

As buyers ask AI assistants instead of typing keywords, discovery is shifting. What marketers should do now — from GEO to first-party measurement.