91% of Marketing Teams Use AI. Only 41% Can Do the Math.

A while back I read a report, and the further I got, the harder it was to sit still.

Jasper and Benchmarkit — one a marketing agent platform, the other an independent third-party benchmark research firm — teamed up to survey more than 1,400 marketers, across industries from tech and finance to retail, media, and life sciences.

So why couldn't I sit still?

Because I realized the AI story in marketing has quietly turned to a new chapter. And a lot of people haven't noticed yet.

What Is the "Operational Era"?

Start with one number: 91%.

That's the share of marketing teams currently using AI. The year before, the number was 63%. In the space of a year, we went from "most people use it" to "practically everyone does."

Even more interesting are the two numbers below.

97% of marketers say whether a company uses AI influences where they choose to work. 75% call it a key criterion when they evaluate a new job.

Let that sink in. AI is no longer a tool. In the eyes of job seekers, it's part of a company's "infrastructure" — like whether the office building has air conditioning.

Jasper has a name for the current stage: "the operational era of AI."

What does "operational" mean?

In 2025, everyone was running experiments: drafting copy, generating images, testing whether AI was any good. In 2026, the experiments are over. AI has moved into the workflows: how a campaign gets planned, how content gets produced, how results get measured — it's in all of it.

Experiments ask "does it work?" Operations ask "how do we scale it?"

These are two completely different questions.

Spending Is Up. The Math Is Getting Harder.

Now let's talk about money.

95% of marketers plan to increase their AI investment this year. 66% intend to put more than 10% of their marketing budget into AI. Among the most aggressive companies, 87% have already done so.

Spending this hard — what about the returns?

Here comes the most painful number in the entire report: only 41% of marketers can confidently prove that their AI investment has produced a return.

The year before, that number was 49%. It didn't go up. It went down.

Is AI no longer working?

No. The passing bar has been raised.

It used to be enough to tell your boss: AI saved me 20 hours. The boss nods — not bad. Now that AI has entered the core workflows, what the boss wants to see is: how many more leads? How many percentage points of revenue contribution? How much better did the campaigns perform?

"Hours saved" doesn't buy "revenue earned."

The math didn't get harder — the ledger got upgraded.

And the people who can do this math are eating well. Among teams that seriously track AI returns, 60% have achieved at least a 2x return. Among high-maturity companies, 61% can produce proof of ROI; early-stage companies don't even come close.

In the same wave of AI, some people are picking up seashells while others are hauling out gold bars. The difference is whether you keep that ledger.

The New Bottleneck: Process

Last year's pain points? Not enough budget, not enough skills.

This year the pain has moved. It's now "governance."

In plain terms: AI can churn out a pile of content in seconds, but the company's review processes are still designed around "weeks." Legal has to sign off, brand has to sign off, compliance has to sign off too. By the time all the stamps are in place, the trending topic has already gone cold.

Concern about legal, compliance, and brand review jumped 3.4x in a single year — the fastest-rising item in the whole survey.

The machines are already flying down the highway, but the toll booths still have human clerks.

This is the real hurdle of the operational era. It's not that companies can't afford the tools — it's that their processes can't keep up.

Bosses and Employees Live in Two Different Worlds

There's another set of numbers in the report that I stared at for a long time.

61% of CMOs say they are confident in the return on AI.

And frontline employees?

12%.

One sits at the top of the org; the other is in the trenches — and they're watching two different wars. The boss sees a strategic opportunity in a slide deck. The front line is carrying the work with no training, no guidelines, and no role boundaries.

Organizations are reshaping themselves accordingly. For one in three marketers, AI is now formally written into the job description: writing prompts, building workflows, setting standards, making strategy. 65% of companies have created positions dedicated to managing AI workflows.

Over the next 12 months, the three kinds of people companies most want to hire:

AI search specialists — 40% of companies plan to hire them, to study visibility in AI-native search. AI transformation leads — 34%, owning cross-functional strategy and change. AI architects — 31%, owning the technical foundation and systems integration.

Look at these job titles. Which of them belonged in a marketing department three years ago?

Three Stages

So what kind of company counts as having "made it"?

The report sorts companies into three stages.

Stage one, experimentation: employees all use their own tools, with no standards. Stage two, sanctioned use: teams have approved tools, but little method to them. Stage three, operations: AI is embedded in formal processes — with governance, with ownership, and with platform integration.

The companies that make it to stage three have strikingly similar playbooks:

They build content as a system rather than treat it as one-off tasks. They embed governance directly into workflows rather than bolting on review after the fact. They assign every piece of AI output to a specific owner. And when they do the math, they don't just count the hours saved.

One more thing: 83% of respondents say their company's leadership is seriously committed to AI. Among the most mature companies, 86% describe that commitment as "very strong." The beginners? 32%.

AI is the boss's business before it is everyone else's. How seriously the person at the top takes it almost single-handedly determines how far you climb.

But Every Story Has a Flip Side

All of that is the view from inside the marketing industry. Steam rising everywhere, everyone charging ahead.

Now let's step outside the marketing department and look at the wider water. There are several chunks of ice floating out there.

The first chunk comes from venture capitalist Tomasz Tunguz. He analyzed 374 quarters of data from 25 public software companies, tracking one metric: net dollar retention, or NDR.

What the metric means: of the money your existing customers paid you last year, what percentage are they paying you this year? Above 100% means existing customers are buying more and more.

The past few years saw a slow decline: from 125% in 2022 to 112% in 2025. No cliff. Not on the day ChatGPT launched, and not when enterprises bought Copilot licenses en masse either.

Then 2026 arrived.

Within a single quarter, the bottom 25% of companies in the industry saw NDR drop from 106% straight to 101% — brushing right up against the break-even line.

Company by company: Zoom at 98%, Asana at 96%. Existing customers are pulling back.

The common thread among the companies that fell behind, as Tunguz puts it bluntly: their products are simple enough to be replaced.

What AI eats first is the business that is easy to replace.

The second chunk comes from a labor report by Anthropic. It proposes a new metric called "observed exposure": it doesn't just look at what models could theoretically do, but at what people actually use them for — and automation-type uses carry a higher weight.

Two findings worth remembering.

First, AI is far from its theoretical ceiling. Take "computer and mathematical" tasks: what Claude actually covers today is just 33%. The blue ocean is still enormous.

Second, the practitioners with the highest exposure tend to be older, more often women, more educated, and better paid. And since the end of 2022, these high-risk occupations have not seen a wave of unemployment. What has actually slowed is the hiring of young people.

The veterans haven't been replaced yet — but the newcomers can't get in the door.

The third chunk is Anthropic's own business. As of January, its revenue over the trailing twelve months had reached $19 billion. OpenAI's was $25 billion. The gap is narrowing, but what's truly startling is the enterprise chat market: in January 2025, Anthropic's share had just cleared 10%, and OpenAI's was close to 90%. Now, that number is almost 70%.

In eighteen months, the whole picture turned upside down.

Enterprises can comparison-shop for models, but more and more of them are handing their first dollar to Anthropic.

A Bucket of Cold Water

Finally, let me introduce someone: Ed Zitron. Silicon Valley's famously sharp-tongued AI skeptic, who runs his own tech PR firm. He wrote a widely circulated long essay whose core argument fits in one sentence:

This AI boom is a bubble propped up by financial engineering, credulous media, and deliberate misdirection by the big players.

From his evidence, let me pick out a few of the hardest pieces.

Nvidia's latest quarterly earnings beat expectations, but buried inside was $27 billion in "cloud commitments." To put it bluntly, it is subsidizing its own customers to rent its own chips. What does that suggest? Real demand isn't as strong as it looks.

CoreWeave — the "neocloud" company Nvidia holds a stake in — posted a quarterly loss. Revenue per megawatt of compute is falling; the more it expands, the less each unit is worth. 67% of its revenue is tied to Microsoft alone, and its capital spending in 2026 is set to double to $20 billion. The business of renting out GPUs may be structurally unprofitable.

And that headline-making "$110 billion OpenAI funding round"? When Zitron took it apart: of Amazon's promised $50 billion, only $15 billion is actually committed, with the rest contingent on AGI or an IPO; Nvidia wrote in its 10-K that the deal is still being "finalized" and there is "no assurance" it closes; SoftBank is paying in quarterly installments, with even that money depending on bridge loans.

OpenAI's own forecast has drifted from "$44 billion in losses by 2028" all the way to "$230 billion in losses by 2030." Zitron ran the numbers himself, and his conclusion is that the actual losses are almost certainly far greater than what has been publicly admitted.

In his view, OpenAI and Anthropic are fighting an information war: feeding credulous media, manufacturing confidence among investors and the public. Revenue charts deliberately make people read "annualized figures" as "actual revenue." Capabilities are systematically overstated.

The most brutal passage is about "principles."

The Pentagon at one point wanted to list Anthropic as a supply-chain risk, and Anthropic refused. Many people praised it for having a backbone. Zitron dug up its own statement: it supports "all legal uses" of AI for national security, excluding only mass domestic surveillance and fully autonomous weapons.

Meanwhile, Claude has long been doing intelligence analysis and target identification for US Central Command, and was used in the Iran conflict. Anthropic has also pitched its technology at the Pentagon's drone-swarm competition. Its entire reservation about autonomous weapons is that "the models are not yet reliable enough" — and it has offered to help improve them.

And Sam Altman? Seizing the opening his rival left behind, he signed a contract with the Pentagon, with the public explanation being "red-line equivalent." The core phrase of the contract is likewise "all legal uses." Under existing legal precedents, including Section 702 of FISA, that is roomy enough to accommodate mass domestic data collection. The Pentagon subsequently confirmed: its position on surveillance has not changed.

Zitron's conclusion is hard to listen to, but it deserves to be relayed as is. These two companies, while being celebrated as visionary thought leaders, are quietly turning war into business; while claiming their models can replace white-collar workers, they use "not reliable enough" as a shield to dodge responsibility for their use in the military.

And the lies holding up this bubble — fabricated revenue, overstated capabilities, forecasts reported as fact — have ultimately led the military to trust a technology that hallucinates, using chatbot output to validate decisions with lethal consequences.

The sellers have been lying all along, and the buyers deserve to know it.

Back to That Report

All right. Now that the cold water has been thrown, let's return to the survey of 1,400 marketers we started with.

On one side: the marketing industry's real-money investment, new roles sprouting like bamboo shoots after rain, ledgers growing thicker by the day. On the other side: hairline cracks at the foundation, sharp skepticism, and warnings that it "might not hold up."

Both of these pictures are true.

There's a line at the end of the report that I very much agree with: the winners will not be the companies with the most AI tools, but the companies that redesign accountability, standards, and measurement into every layer of their processes.

Anyone can afford the tools. The process is the moat.

As for whether the bubble will burst, and when — I don't know. But I do know one thing: no matter how far the tide goes out, the person who can do the math is always the last one to panic.

May you be that person.