OpenAI’s November 10 study of freelance work offers a more useful marketing question than whether AI will eliminate an entire profession: which parts of a client’s project can they now attempt themselves, and which parts do they still need someone to deliver? The answer can change the shape of an agency offer even when the underlying business objective stays the same.
The published analysis compares ChatGPT activity with Upwork marketplace postings. These are different populations, not a matched experiment following the same clients from AI use to hiring. Read the findings as descriptive evidence about work composition. The interpretation below separates the study’s results from practical suggestions for marketing teams; the primary report and accompanying article were checked on November 17, 2026.
- 01Share is not headcountA category can lose marketplace share while its number of postings rises. Neither measure alone tells you how many people lost jobs.
- 02The pattern is mixedThe study does not support a simple claim that every activity common in ChatGPT is disappearing from freelance work.
- 03Scope changes the budgetA larger fixed-price assignment may include more work. Do not turn a project-budget difference into a universal hourly-rate premium.
- 04Review the deliverableMap the complete client outcome, including integration and quality control, before changing a service’s price or staffing.
01 — Study boundaryRead the 63-category comparison correctly
Across 63 matched work subcategories, 43 followed the overall pattern: activities relatively common in ChatGPT tended to lose Upwork share, while less-common activities tended to gain share. The other 20 went in the opposite directions. This does not mean that 43 categories all declined. The scatter chart compares ChatGPT category shares in the second half of 2025 with Upwork’s first-half-2023 baseline, and changes in Upwork share from that baseline to the second half of 2025.
The published article also contains a different overview using ChatGPT activity from January 2025 through June 2026 and Upwork comparisons between the first halves of 2025 and 2026. Do not combine the charts into one universal study window. When a number matters to a decision, preserve its chart, population and period. A category trend over one interval is not automatically evidence about the most recent quarter.
For a marketing manager, the useful starting point is an inventory of services rather than a prediction about occupations. Copy drafting, video production, campaign configuration and reporting can appear within the same client engagement while responding differently to AI tools. A category-level association helps identify questions to investigate. It does not determine what a particular client will buy next, whether a specialist is becoming unnecessary or how a specific team should be reorganised.
The study compares patterns in two sources of activity. It does not follow a single customer from using ChatGPT to cancelling a freelance contract.
02 — Share arithmeticDo not turn a changing mix into a job-loss count
Marketplace share describes a category’s proportion of the measured total. It can change because that category changes, because the rest of the marketplace changes, or both. An illustrative example makes the distinction clear: a category grows from 100 to 120 postings while total postings grow from 1,000 to 2,000. Its volume rises by 20%, yet its share falls from 10% to 6%. Neither number is wrong; they answer different questions.
The same care applies when a report discusses tasks inside postings. A project can request several tasks, so the share of task assignments is not the same denominator as the share of job posts. A posting also is not automatically a completed hire, a full-time role or a particular person’s income. Keep the unit of analysis visible before converting a research chart into a staffing or revenue forecast.
For your own business, examine actual briefs, accepted projects, realised revenue and delivery effort alongside the mix of requests. If standalone copywriting becomes a smaller share of new enquiries, check whether its absolute volume fell and whether writing has moved inside larger packages. Those possibilities require different responses. A changing label in a proposal should not be mistaken for the disappearance of the work itself.
| Illustrative period | Category postings | All postings | Category share |
|---|---|---|---|
| Earlier | 100 | 1,000 | 10% |
| Later | 120 | 2,000 | 6% |
03 — Marketing tasksLook for the work that connects an output to a result
The full OpenAI–Upwork report identifies gains in task share within AI-related postings between 2023 and 2025 for video production, social-media campaign management, marketing automation and AI API integration. These are changes in the mix of requested tasks, not totals for all marketing employment. They give a concrete reason to examine how production and implementation fit together in your service catalogue.
Consider a hypothetical client who can generate draft email copy but needs a functioning lifecycle campaign. The remaining work may include deciding the audience rules, connecting the right customer events, checking exclusions, reviewing the messages and verifying the reporting. The draft is one component of the delivery. A proposal that describes only a number of emails can leave those responsibilities unclear, even though they determine whether the campaign is usable.
Our marketing automation workflow guide provides a way to map those connections. Use the research as a prompt to inspect your own briefs, not as proof that every client wants a larger automation project. Some customers still need a narrow specialist task. The service design should match the missing capability and the accepted deliverable, rather than adding AI terminology to work that has not materially changed.
Create the component
Specify the draft, video, design or analysis and the quality standard it must meet.
Connect the workflow
Identify the systems, approvals and data needed to put the component into use.
Verify the outcome
Define who checks the finished work and which evidence demonstrates that the agreed scope is complete.
04 — Budget evidenceSeparate bigger projects from higher hourly prices
The report’s comparisons through April 2026 find adjusted posted fixed-price budgets about 52% higher for AI-related work, but adjusted posted hourly rates only about 1.5% higher. Hours worked in the first 90 days were about 41% higher among hired hourly contracts with positive recorded hours. The adjustments control for work subcategory and posting month. These are associations, not an identified causal return from adding AI to a service.
The distinction matters when pricing a marketing engagement. A larger project budget may purchase a broader scope, more rounds of refinement or more implementation work. It does not show that a freelancer can charge every client an additional 52% for the same deliverable. Posted budgets also differ from realised revenue and profit. An attractive proposal value can conceal substantial coordination, revision and support effort.
Use the agency pricing decision guide to match the commercial structure to the work. If the scope is predictable, a fixed deliverable may be appropriate. If discovery is substantial, make that uncertainty visible and define the decision after discovery. If ongoing optimisation is included, specify what is monitored and changed. Price the responsibility you accept and the evidence you will deliver, rather than borrowing a research percentage as a rate card.
Before changing a fee, compare the agreed scope, revision allowance, implementation effort and support obligation. A larger assignment is not the same product at a higher price.
05 — Service designPackage a complete outcome without hiding the boundaries
A useful service redesign starts with a deliverable the client can recognise. For example, a hypothetical video campaign package might include a concept, a defined number of edited assets, approved formats and a measurement plan. AI may assist parts of production, but the client still needs to know what will arrive, who approves it and which responsibilities remain with their team. Avoid selling an undefined promise of greater output.
Write down the handoffs that usually cause delay. Does the client need to supply product information, approve claims or provide access to a publishing account? Does a designer need final copy before editing begins? Does reporting depend on events that are not yet configured? Those dependencies belong in the proposal because they affect delivery. Combining tasks can reduce friction only when the combined responsibility is genuinely understood.
The content production and agency margins guide examines why more output alone is an incomplete operating goal. A team may generate additional drafts while spending more time deciding which ones are usable. An offer should therefore include selection and quality control where those are part of the client’s need. Keep optional work visible, so a broader package does not become a mechanism for selling unnecessary scope.
| Offer element | Question to resolve | Evidence of completion |
|---|---|---|
| Asset production | Which formats and quality standard? | Accepted deliverables |
| System connection | Which integrations are included? | A verified end-to-end handoff |
| Client approval | Who decides and by when? | A recorded acceptance decision |
| Ongoing work | What happens after launch? | Defined monitoring and change scope |
06 — Team capabilityAllocate work by judgement and responsibility
The research should not be used as a shortcut to a headcount decision. Start with the tasks needed to fulfil the offer and the evidence that each can be performed reliably. A junior marketer may use AI to prepare options while a specialist checks product claims and an account lead resolves conflicting client priorities. The useful unit of design is the handoff between those responsibilities, not a blanket label that a role is automated.
Choose where human review is essential to the promised result. A factual claim, a sensitive audience choice or a material budget change needs an accountable decision. Routine formatting may require a lighter check. The review should be specific enough that the reviewer knows what to inspect and what would cause rejection. Asking someone to “check the AI output” without a standard creates work without a clear acceptance boundary.
Training should follow the same task map. If the bottleneck is inaccurate customer data, prompt-writing practice alone will not resolve it. If the bottleneck is weak creative judgement, a faster integration will not make the work persuasive. Invest in the capability that limits the deliverable, then assess whether the change improves accepted work. This approach lets a team benefit from AI tools while preserving the skills needed to recognise an unsuitable result.
07 — Operating measuresMeasure accepted work and the effort behind it
Create a small operating view that distinguishes production speed from delivery success. Count accepted outputs, revision cycles, time spent on review and the proportion of projects completed within the agreed scope. Where business outcomes are part of the engagement, define how they will be measured and what other factors affect them. Do not credit every movement in campaign performance to the introduction of an AI tool.
Keep the denominator honest. If a team produces ten drafts and accepts two, its useful output is not ten completed assets. If a campaign package grows in revenue but also requires much more support, the revenue change alone does not establish better margins. Track unsuccessful attempts and rework as part of the effort required to deliver the accepted result. That makes workflow improvements visible without presenting internal activity as client value.
Compare similar work where possible and document what changed. A pilot serving a different client segment or adding more deliverables is not directly comparable with the old offer. The content team productivity metrics guide provides a starting set of questions. Use a consistent definition of completion, and retain enough context to explain why a faster process did or did not produce a better delivery outcome.
Measure the effort required per accepted deliverable, including rejected drafts, review and rework. Output volume by itself can hide the cost of deciding what is usable.
08 — Practical pilotTest one revised offer before rebuilding the agency
Choose a service where you already understand the client need and the current delivery process. Write a revised scope that connects the production task to its implementation and acceptance criteria. Identify which AI-assisted steps are expected to help and where an accountable person remains responsible. This creates a testable proposal rather than a claim that the entire agency has become more efficient.
Run the offer on a manageable number of suitable engagements and review the actual work produced. Record client questions, missing inputs, review effort and any ambiguity over what was included. If clients value the integration but not the extra asset volume, adjust the package accordingly. If the broader offer creates coordination work that erases the apparent gain, narrow the scope or change the handoff before expanding it.
Our content engine service connects editorial planning, production and distribution around an agreed business need. The study is useful because it encourages that complete-delivery view, while leaving the commercial decision to evidence from your own customers. Use the market pattern to choose a question, then let accepted work, client feedback and delivery economics determine the answer. That is a more reliable basis for changing an offer than extrapolating a category chart into a universal forecast.
Use the study to improve a service, not predict a profession
OpenAI and Upwork’s research supports a closer look at how marketing tasks are combined and purchased. It does not provide a simple job-loss count or a guaranteed AI price premium. Review the whole client deliverable, make its boundaries clear and test whether a revised workflow improves accepted work and delivery economics. Those are decisions a marketing team can evaluate directly.