©2026 ZenABM - All Rights Reserved.
An AI LinkedIn ads audit in 2026 is not about using an AI wrapper that gives generic advice.
It is a set of prompts you run against your own account, and this post gives you every one of them.
When I ran a 7-figure ABM program on LinkedIn, the audit was the job I put off most.
The data pulls ate a whole day before I got to think.
Now the AI does the pulls, the sorting, and the first pass at what looks wrong.
I spend my time on the decisions instead.
After all, a big part of successful marketing is simply catching waste early.
Below is the full prompt library to do that from your terminal.
You also get the plain-English math behind each check, so you know what the AI is computing, plus a scored checklist you can run this week.
Paste the prompts one by one, or save them as one skill and run the whole audit every month.
The short version:
/linkedin-abm-audit skill packages the whole thing so you can skip prompts ans simply run the skill to catch the problems in time.
Do not think of the AI as the complete auditor.
It is not.
It replaces the slow part: the exports, the sorting, the cross-checking, and the first look at what seems off.
That is the part that made audits take a day, and the part teams keep skipping.
What stays with you is the judgment: does this finding matter for your ICP, your stage, and your goal?
So the rule for every prompt below is simple.
The AI reads and proposes.
It does not change anything.
Later, you might ask it to act on a finding, like pausing an ad.
That becomes a write action, and a good connector asks you to confirm before it runs.
You are never one prompt away from an accident.
For the full manual version of these checks, and the step-by-step logic behind each score, read our LinkedIn ads audit walkthrough.
This post is the AI version: the same checks, run as prompts.
Every prompt below assumes the AI can see your account.
There is no official LinkedIn connector for Claude or ChatGPT.
So the link goes through a third-party MCP (Model Context Protocol) server.
It reads the LinkedIn Ads API and joins it to your CRM.
With ZenABM, the path is short.
Start an account, connect LinkedIn Ads and your CRM, then add the server at https://app.zenabm.com/api/mcp to Claude Code over OAuth or an API token.
Run /init so the AI writes itself a note about how your account is built.
After that, you just ask.
Here is the part you must not skip: the LinkedIn Reporting API does not hand you everything by default.
It returns only impressions and clicks unless you ask for more, and you can request up to 20 metrics at a time.
Its demographic numbers are approximate on purpose, within about three of the real count.
It also keeps that demographic data for only two years.
So the quality of your audit depends on the connector, not the AI model.
The connector has to pull the right fields and join them to your CRM.
Two metrics matter most, and Campaign Manager does not show them cleanly:
ZenABM works in eCTR and eCPC.
It pulls its engagement data straight from the LinkedIn API at the company level, so the audit ends in pipeline, not vanity numbers.
The full connection steps are in the guide on building ABM workflows with Claude Code.
Connected the ZenABM MCP server to your terminal?
Below are the seven checks you can run during every audit now.
Each one has a prompt to paste and a short note on the math it runs, so you know what to trust in the answer.
Treat the numbers in the prompts as starting points and change the thresholds to fit your account.
An audit that skips tracking is auditing numbers it cannot trust.
Start by checking that the Insight Tag fires, conversions are defined right, UTMs are consistent, and the CRM sync is intact.
The AI can match the conversions in the account against the events landing in the CRM and flag where they split.
Check my conversion tracking. List every conversion action, whether it has fired in the last 14 days, and any campaign optimizing for a conversion that has recorded zero events this month. Grade each tracked event as working, partially working, or broken.
What you are hunting for is the silent break.
A conversion that stopped three weeks ago does not send a warning, but every campaign aiming at it has wasted budget since.
If any event grades below working, treat the rest of the audit as directional until you fix it.
This is the one check where you do not proceed on a red flag; you stop and repair it first.
Messy structure hides waste.
Ask the AI to map the account: campaigns by objective, ad sets by audience, and any old campaign still spending that no one remembers.
Inconsistent names are the tell that testing logic has drifted.
Map my account structure. Group campaigns by objective and ad format. Flag any campaign with spend in the last 30 days that has not been edited in 90 days, and list every naming inconsistency you find.
The orphans are the point.
A campaign still spending months after its last edit is usually a test someone forgot to switch off.
It is also the easiest budget to win back.
AJ Wilcox, founder of the LinkedIn ads agency B2Linked, makes the case that naming is not cosmetic.
When you name campaigns after the audience, he argues, you can break down all of your performance by who you reached.
That is how you learn which segments actually work.
If the AI cannot group your campaigns cleanly, that messy naming is itself a finding.
This is where targeting discipline pays off.
Audiences should sit in the 30,000 to 100,000 range.
Exclusions should cover job seekers, competitors, and current customers.
And two ad sets should not be bidding against each other with the same audience.
The AI can also show which job titles got spend that are not on your ICP list.
That waste stays invisible until someone pulls it.
Audit my targeting. List ad sets with an audience under 30,000 or over 100,000, any non-ICP job titles I spent more than $200 on this month, the companies eating the most impressions without engaging, and any two ad sets whose audiences overlap by more than 25%.
Two settings leak budget quietly, so check them by name.
Audience Expansion lets LinkedIn serve your ads to profiles it thinks are similar, which waters down precise targeting.
The LinkedIn Audience Network pushes your ads to apps and sites off LinkedIn, where they behave nothing like the in-feed audience.
For a tight ABM program, both settings should be off.
Add a prompt to confirm their status and show the spend running through them, so the call is made on numbers, not defaults.
List every campaign with Audience Expansion on and every campaign delivering on the LinkedIn Audience Network. For each, show the spend, impressions, and engagement from the expanded or off-network delivery, so I can decide what to switch off.
Exclusions do more than they appear to.
Tim Davidson, founder of B2B Rizz, shared that adding a job-seekers exclusion cut one campaign’s audience by 28%.
That means more than a quarter of that spend had been reaching people who were not buyers.
He also audited an account that targeted 351 accounts but went fully broad, instead of uploading the account list with job titles.
The ABM intent never reached the setup.
ZenABM shows this at the company level.
It names which accounts absorb impressions and where they sit in your funnel, so the targeting audit works on real accounts, not audience averages.
Also, ZenABM lets you exclude a company from campaigns with a single click from ZenABM’s UI:
Most audits get sloppy here by comparing formats that do different jobs.
A Thought Leader Ad and a single image ad are not in the same race, so benchmark each format against itself.
Here are the ZenABM median benchmarks to hold each format to:
| Format | Median CTR | Median CPC |
|---|---|---|
| Thought Leader Ads | 2.68% | $2.29 |
| Single image ads | 0.42% | $13.23 |
| Carousel ads | 0.32% | $13.30 |
| Video ads | 0.24% | $15.61 |
The gap is huge, which is the whole point: a 0.42% single image ad is at benchmark, not failing, and a Thought Leader Ad is 77% cheaper per landing page click than standard formats.
You can access the full ZenABM ABM Benchmarks 2026 Report here.
Ask the AI to rank ads within each format, then identify the ones that are sliding.
For each ad format separately, rank ads by eCTR and eCPC against the format median. Then give me a decaying-ads report: ads whose eCTR has fallen for two or more weeks in a row, counting only ads with more than 1,000 impressions.
I ran similar prompts for our program at ZenABM:
Read the decaying-ads list first.
Creative fatigue on LinkedIn is slow.
An ad slides for weeks before anyone notices, and by the time it shows up in a monthly dashboard, it has already wasted a month of budget.
The math behind the flag is deliberately strict.
An ad only counts as decaying after 1,000+ impressions and two straight weeks of falling eCTR, so you are acting on a trend, not one noisy day.
For Thought Leader Ads, add one line to break out performance by author.
The same message lands very differently depending on whose face carries it.
LinkedIn offers maximum delivery (automated), cost cap, and manual bidding, and the right one depends on the objective.
The trap is that a bid set against the wrong charging model, paying per impression when you want landing page clicks, inflates cost quietly.
So the audit should line up each campaign’s objective, bid strategy, bid amount, and charging basis, then flag the mismatches.
For every active campaign, list the objective, bid strategy (maximum delivery, cost cap, or manual), the bid amount, and what that bid charges for (impression, landing page click, engagement, send, or video view). Flag any manual bid priced more than 20% above the median for its ad format, and any bid strategy that does not fit the campaign objective.
Then check pacing and schedule, because a campaign with no end date is the most common silent leak of all.
Per campaign, show the daily and lifetime budget, how spend paced over the last 30 days (even, front-loaded, or under-delivering), any campaign whose budget is too small to serve its audience, and any campaign with no end date that has spent for more than 60 days.
This is the check that pays for the audit.
Ask the AI to show spend concentration by format, audience, and funnel stage, the ad sets with rising spend and falling engagement, and the accounts eating impressions without ever clicking.
Each of those is budget you can move.
Show me wasted spend: ad sets with the most spend and lowest eCTR, ad sets with rising spend and falling engagement over the last 4 weeks, and the 10 companies with the most impressions and the least engagement.
I ran a similar prompt for ZenABM:
ZenABM scores each account by current versus total engagement, so the waste signal comes with account-level context instead of just an aggregate number, and that score is available in your AI terminal through the MCP.
Naming the impression hogs is what turns “we are wasting spend” into a list you can act on this afternoon.
A B2B audit that ends at lead counts is not finished.
The real question is which campaigns touched the deals that opened and closed.
Because the ZenABM MCP connector joins ad data to the CRM, the AI can answer it directly instead of guessing.
Which campaigns and ad sets were the most common touchpoints before open deals and before closed-won deals? And which spend produced no pipeline at all this quarter?
The answer reframes everything.
A campaign that looks weak on CTR but shows up before half your closed-won deals is not one to cut. It is one to defend.
A campaign with a great CTR that never appears before a deal is the opposite.
Without the CRM join, those two look identical.
That is why an audit that stops at engagement so often recommends cutting the wrong thing.
ZenABM does this join natively against HubSpot or Salesforce, so the pipeline view comes back without a manual export and merge.
Once the seven prompts are done, turn the findings into one scorecard.
Do not invent a precise number like 42 out of 100 unless your scoring is truly that exact.
A three-tier grade per area is more honest and more useful.
Ask the AI to fill it in from what it just found.
Using everything from the seven checks, grade each area Strong, Mixed, or Weak, with one sentence of evidence each: tracking, account structure, audience and targeting, creative by format, bidding and budget, wasted spend, and attribution readiness.
| Area | What to grade | Grade |
|---|---|---|
| Tracking | Insight Tag, conversions, UTMs, CRM sync | Strong / Mixed / Weak |
| Account structure | Naming, segmentation, orphan campaigns | Strong / Mixed / Weak |
| Audience and targeting | Sizes, exclusions, overlap, ICP drift, expansion, network | Strong / Mixed / Weak |
| Creative by format | Each format vs its own benchmark, decaying ads | Strong / Mixed / Weak |
| Bidding and budget | Bid strategy fit, pacing, schedule, end dates | Strong / Mixed / Weak |
| Wasted spend | Spend concentration, impression hogs, rising spend and falling engagement | Strong / Mixed / Weak |
| Attribution readiness | CRM join, pipeline touchpoints, spend with no pipeline | Strong / Mixed / Weak |
An audit is only worth the actions it starts.
Work the findings in this order to put money back to work fastest:
None of these requires the AI to touch your account.
It proposes, you decide, and only when you approve an action like pausing an ad does the connector treat it as a write and ask for confirmation.
The audit never becomes an accident.
Running seven prompts by hand is fine once.
To run the audit every month, save it as a skill so it fires the whole sequence at once.
There are two free options.
The /linkedin-abm-audit skill is one of four free ZenABM Claude skills at ZENABM/linkedin-abm-skills.
It packages this whole audit as a 30-day diagnostic.
It produces a scorecard, checks whether you are running more ads than your budget can fund, grades each format against the benchmark medians, lists decaying ads and impression hogs, and hands back a prioritized fix list as branded HTML and PDF.
Its reporting counterpart, /linkedin-abm-report, turns the same data into an exec recap.
Install both, plus the strategy and ad-design skills, in Claude Code with:
/plugin marketplace add ZENABM/linkedin-abm-skills
/plugin install linkedin-abm-skills@zenabmOn Claude Desktop or web, download the zips from the releases page and upload them under Customize, then Skills.
They run on sample data out of the box, and pull your real numbers through the MCP server once connected.
If you would rather not open a terminal at all, use Zena, the analyst agent inside ZenABM.
Its proactive layer runs the ads decay report and the waste flags on a schedule and shows them on your dashboard, so the audit’s findings arrive before you ask for them.
The other free option is Claude Ads, an open-source Claude Code skill by Daniel Agrici.
It runs 25 automated checks across tracking, audience, creative, lead gen, and budget.
It produces a weighted health score and flags wasted spend and misallocation, all locally, so no account data leaves your machine. It works from exported Campaign Manager CSVs rather than a live connector.
That makes it a strong second opinion, and a good starting point if you are not connected yet.
A LinkedIn ads audit only helps if it actually gets run.
The reason most teams run it quarterly is that the mechanical work used to take a day.
Connect your account to Claude Code through the ZenABM MCP server, and that work collapses to minutes, so the audit fits into a monthly rhythm.
That change matters more than any single prompt.
A broken conversion, an inverted budget, or an ad decaying for three straight weeks all get caught in week five instead of compounding for a quarter.
So paste the seven prompts this week.
Grade the accounts, fix the decaying ads and the impression hogs first, and then save the whole thing as a skill so next month it runs itself (or just use the ZenABM’s skills package).
You can start using the ZenABM MCP server for free with ZenABM’s 37-day trial or book a demo with us to know more!
Yes. With your account connected through an MCP server, Claude Code can run the mechanical part of a LinkedIn ads audit in minutes: pulling every campaign, ranking ads by eCTR and eCPC, detecting decaying ads, and finding wasted spend. It reads and proposes only. The strategic calls about what to change stay with you, and any action like pausing an ad needs your confirmation before it runs.
Seven areas: conversion tracking, account structure and naming, audience and targeting, creative performance by format, bidding and budget, wasted spend, and attribution to pipeline. The connector adds metrics Campaign Manager does not show cleanly, like eCTR (the effective click-through rate to your landing page) and eCPC (the cost of each of those clicks), and it joins ad spend to CRM deals.
Yes. Every audit check is a read, so nothing in the account changes while it runs. If you later ask the AI to act on a finding, such as pausing an ad or excluding a company, that is a write action, and a good connector like the ZenABM MCP server asks for explicit confirmation before it happens. You cannot trigger an accidental change with an audit prompt.
Use format-specific benchmarks, not generic averages. From the ZenABM 2026 report: Thought Leader Ads run a 2.68% median CTR at a $2.29 CPC, single image ads 0.42% at $13.23, carousel ads 0.32% at $13.30, and video ads 0.24% at $15.61. Compare each campaign to its own format. A 0.42% single image ad is at benchmark, not failing.
Monthly, once it is automated. LinkedIn ad performance decays slowly, so a quarterly audit catches waste three months late. Saving the seven checks as a repeatable skill, like the free /linkedin-abm-audit, lets it run on the first of every month with findings you can compare over time. A lightweight weekly check of tracking health and winner and loser patterns fills the gaps.
The data work that used to take most of a day takes a few minutes once your account is connected, because the AI reads every campaign in one pass instead of waiting on exports. The rest is judgment: deciding which findings matter. In practice a first full audit takes about an hour of attention, and monthly re-runs are shorter because only what changed needs a fresh look.