The Review Tax: How AI Workslop Lands on Your Senior Consultants

AI made the first draft of consulting work cheap – but every claim in that draft still has to be checked before it goes to a client, and in many firms the person checking is a senior manager or a partner. So the cost of the work moved from making it to checking it, and much of it landed on the most expensive hours in the firm. A Stanford Social Media Lab and BetterUp survey on AI "workslop" put a number on it: 40% of 1,150 full-time US desk workers said they had received it in the last month (BetterUp), and each case took an average of 1 hour and 56 minutes to deal with (Axios).

Partners we speak with call it a verification tax. I call it the review tax. My take is that the fix is to move the checks. Sources, citations and the firm's quality bar get checked before a senior opens the draft, so the senior spends the hour on what the client should do.

Where the review tax comes from

Volume went up – a junior with Copilot or ChatGPT can turn out a long market scan in an afternoon – and trust went down, because the text reads well whether it is right or not. Plausible-but-wrong is slower to check than to write. A wrong number in a fluent draft looks like every other number around it, so the reviewer has to open the sources to find the one that doesn't hold. Leslie Willcocks of LSE calls it the reverse of Polanyi's paradox: a model can tell more than it knows. Fluency makes verification more valuable.

When 22 public service broadcasters tested more than 3,000 responses to news questions from four widely used AI assistants, 45% had at least one significant issue and 31% had serious sourcing problems (EBU). In our own work, on questions with timeframes in a narrow field, citation hallucination can be up to 50% without a citation-checking step (FifthRow). Not super surprising.

What it costs a consulting firm

The junior saves hours on the draft. The senior spends hours checking it – and those are the hours that cost the most.

In a firm that sells days, that's where the margin goes. A partner's evening spent opening an analyst's sources isn't billed at the partner rate – usually it isn't billed at all – while the client is already asking for a lower day rate because AI is in the work.

On calls with consultancy partners the same picture keeps coming up. Clients expect lower fees because they know AI is in the work. The firm's own AI bills go up. Seniors end up fact-checking drafts that juniors produced with AI, and they are tired of it.

And it lands on a business with less slack. Billable utilization in professional services fell to 66.4% in 2025, the lowest in SPI Research's survey history (Deltek).

AI didn't remove the bottleneck. It moved it downstream, and in a consultancy downstream is usually the partner's review.

What happens when the check fails

In October 2025 Deloitte Australia agreed to repay the final instalment of its A$440,000 contract for a report to Australia's Department of Employment and Workplace Relations. A university researcher had found a fabricated quote from a federal court judgment and references to academic papers that don't exist. The revised report disclosed that Azure OpenAI was used in writing it (AP).

I don't read this as a story about one firm. Most firms rely on the same control – a busy senior reading the final draft – and it works until a reference looks plausible and the deadline is close.

How to avoid AI workslop: move the checks upstream

Order matters here. An analyst's wrong number doesn't stay in the appendix – it flows into the market sizing, then into the business case, then into the recommendation. As I put it on a partner call in November: "If they happen at the front of the process, you'll be somewhere completely different at the end of the process". Most of the checks that catch a fake citation are mechanical, so they don't need a partner – they need to run on every claim before anyone reads the draft:

• Every claim carries its source from the moment it is written, sentence by sentence, not as a list at the end.
• Every citation is checked twice: the source exists, and it actually says what the draft claims. If it doesn't, the claim is thrown out.
• "What good looks like" is defined upfront, by the firm, and a judge evaluates the draft against that definition – not against vibes.
• Whatever couldn't be verified is marked, so the reviewer sees what is solid and what isn't before reading a word.

The checklist itself is in our guide How to Verify AI Research.

That is what the Large Consulting Model does. We check every verifiable claim in a written answer against the downloaded sources, and if it's unsupported we throw it out. Then a judge checks whether the work is explained properly for an executive, whether the gaps and assumptions in the method are stated, and whether there are "no unsupported claims" – "and only then it goes into a final output". Quality control is a judge agent in FifthRow that reads every output – it checks your quality bar, whether something was left out on purpose, where each source came from, and whether a claim is assumed or triangulated (FifthRow). The quality bar is the firm's own, because the firm puts its own method in. And the team keeps its tools – it runs inside Copilot, ChatGPT and Claude.

I think fact-checking is simply the cost of trust – if we show a partner something that is just wrong, they won't trust the next thing either.

What the senior's review becomes

Once the checks run first, the senior can stop asking "is this true?" and start asking "is this right for this client, and what do we tell them to do?". That second question is what clients pay a partner for. No checker answers it. And it doesn't happen once at the end. "It's never just one thing and it's done" – it's a step, then a person's judgment, then the next step.

What it looks like in practice

A $30B+ global IT services and consulting group runs the research behind its client work on FifthRow and gets citation-backed research back. The practice's own delivery leads put the productivity gain on research at 3 to 5 times, and idea-to-concept work that took weeks of senior consultant time comes back in hours.

At a boutique global brand strategy consultancy with no junior bench, the quality review runs by default before a strategist sees anything.

Frequently asked questions

What does AI workslop mean?

The researchers who named it define workslop as "content that appears polished but lacks real substance, offloading cognitive labor onto coworkers" (HBR). In a consultancy that labor often lands on the senior who signs the work off – and the final reader is a paying client.

How do you avoid creating AI workslop?

Check before you send. Every claim should carry its source, every citation should be confirmed to exist and to say what you claim, and anything unverified should be marked. In a firm, build those checks into the workflow before a senior sees the draft, so avoiding workslop doesn't depend on one tired reviewer.

How much senior time does reviewing AI drafts take?

We haven't found a reliable public figure for consultancies yet. The closest public number is from the workslop research: nearly two hours of rework per instance, self-reported by US desk workers in general, not consultants (HBR). The more useful number is your own: have seniors log review time on AI-assisted drafts for two weeks, split into checking sources and deciding what to recommend.

What I'd do about it

So the practical takeaway for a firm that uses AI in delivery:

• measure how much senior review time goes on checking sources
• write down what good looks like for your core deliverables, so a judge can check against it before a senior does
• don't let anything reach a partner without its sources attached and checked

There are two ways to do this with us: a license for your own delivery, where you run your method on FifthRow, or partner in delivery, where we co-own the work and you keep clear ownership of the client relationship and the quality bar. Either way, you keep the client and the last mile. More on both at fifthrow.com/consultants.

What share of your partners' review time last month went on checking sources rather than on the client? If you've measured it – or want to – book a slot with me and we'll compare notes.