On 13 November 2026, Search Engine Journal published a piece by Greg Jarboe on a survey from CodeSignal, a skills assessment vendor. The survey ran in September 2026 through Pollfish, with 1,000 US employees and 1,000 US people managers. It found that 72% of workers say their AI skills have never been formally evaluated. It also found that 69% of managers say AI ability is a major or moderate factor in hiring.
Jarboe applies this to PPC managers. His advice is to keep a log of Google Ads automation decisions and bring it to your review.
The excitement: AI skill is now a hiring factor
The claim underneath the survey is simple. Employers want AI skills, nobody is checking them, so people need a way to prove theirs. That is a real problem. If 69% of managers weigh AI ability in hiring, and most workers have never been assessed, a hiring manager is mostly guessing.
I would treat the numbers with some care, and Jarboe does too. The 72% is self-reported. CodeSignal sells assessments, so a finding that skills go unassessed suits its business. Jarboe also notes that the launch announcement doesn’t describe the survey method. One more figure cuts the other way. 82% of employees said their manager understands their skills very or somewhat well.
So the gap may be smaller than the headline. But the underlying need holds up. Someone who hires for AI skill has no good way to see it.
What the survey doesn’t fix: a tool list proves little
Look at how people answer today. They say they use ChatGPT, Claude, Gemini, Performance Max, Smart Bidding. Jarboe puts it well: saying you use AI for everything doesn’t prove much.
The survey backs this up. 62% of employees received employer AI training, but only 35% said it’s where they built the skills they use. Most learned from free resources (46%) or trial and error (39%). A tool list or a course certificate tells you what someone was exposed to. It doesn’t tell you whether they can run the tool well.
The tool was never the skill. Tools expire. Judgement does not. The interface you list on a CV will look different in a year. What stays is whether you can tell good output from bad.
The proof is a dated record of your decisions
The habit I teach is this. You direct, the AI produces, you check and decide. You are responsible for every output that leaves your hands. If that habit is real, it leaves traces. You can write them down.
Jarboe’s automation log is exactly that. For each task, record what you let automation handle, what you checked or changed, and the result. Include the recommendations you rejected. Include the negative keywords you added after a broad match expansion. Include the auto-generated assets you paused. Google Ads change history shows what changed, but not why. The log supplies the why.
A good record has three parts:
- A date. AI behaviour changes, so a record without a date is hard to trust.
- An override. Where you disagreed with the tool and what you did instead.
- A result. What happened afterwards, including when you were wrong.
The last part matters most. Someone who has never overridden the tool, or never been wrong, has probably not been checking. Trust nothing unverified applies to the person using the AI as much as to the output.
What this looks like for an operator
Take a retail shop hiring someone to run its Google Ads. Or a clinic choosing a marketing agency. Or a marketing team adding a new person who says they are good with AI.
Asking “which tools do you use?” gets you a list. Ask this instead: “Show me three times you overrode the tool, what you checked, and what happened.” A person with real judgement answers quickly and with detail. They can name the campaign, the recommendation they rejected and the number that moved. A person who only operates the tool tends to go vague.
The same applies to an agency. If they say AI runs your account, ask to see their log. Ask what they rejected last month and why. An agency that checks its automation will have a record. One that doesn’t will talk about its tools.
Where this shows up in my own work
I run my own businesses with AI and check the output before it goes out. My coffee store test is an example. I dated every result and showed where it fell short, because AI answers change week to week. That is the same habit, written down.
What to do
- Ask every hire and every agency for a dated record of overrides, checks and results. Don’t accept a list of tools in its place.
- Ask for rejected recommendations and mistakes, not only wins.
- If you manage a team, keep a simple log yourself first. Task, what the AI handled, what you checked or changed, and the result.
- Judge training by whether people use it on their weekly work, not by whether they finished the course.