Agencies invoice for AI-generated deliverables by listing each output as a line item, separating AI production cost from human review cost, and pricing based on the deliverable’s value to the client rather than the time it took to produce.
A 2025 industry survey found that 61% of agencies now use AI agents for at least some client deliverables. Only 23% have updated their invoicing to reflect this. That 38-point gap is where confusion, disputes, and lost margin live. Clients receive work they did not know was AI-generated. Agencies bill at rates that no longer match their cost structure. Neither side has a clear picture of what was produced, by whom, and at what cost. This guide covers what counts as an AI deliverable, how to format line items, when to separate AI and human costs, how to handle revisions, and how to manage client expectations throughout.
Key Takeaway: 61% of agencies use AI for deliverables, but only 23% have updated their invoicing. Close the gap before clients close it for you.
What Counts as an AI-Generated Deliverable?
The first invoicing challenge is classification. Not everything an AI agent touches is an “AI deliverable.” Drawing the line matters for pricing, client communication, and trust.
Content deliverables. Blog posts, ad copy, email sequences, social media posts, product descriptions, landing page copy. If an AI agent produced the first draft, it is an AI-generated deliverable — even if a human edited it extensively afterward.
Design deliverables. Layout concepts, image variations, banner sets, presentation templates. AI-generated design work is growing fast, particularly for high-volume assets like social media graphics and ad variants.
Code deliverables. Landing pages, email templates, automation scripts, data dashboards, API integrations. AI coding agents can produce functional code that human developers then review and refine.
Research deliverables. Market analysis, competitor audits, audience insights, trend reports, media overviews. AI agents gather, synthesise, and structure research data that human strategists then interpret and contextualise.
The grey area: AI-assisted vs AI-generated. When an AI agent produces a complete first draft and a human makes minor edits, that is AI-generated. When a human creates the strategy and structure while an AI agent fills in supporting data points, that is AI-assisted. The distinction matters. Clients perceive “AI-generated with human editing” differently from “human-created with AI support.”
A practical rule: if removing the AI contribution would require the human to start from scratch, call it AI-generated. If the human work stands on its own and the AI contribution enhanced it, call it AI-assisted. Be consistent across clients.
How Should Agencies Format AI Line Items on Invoices?
Four approaches work. Choose based on your client relationships and billing model.
Approach 1: Deliverable-Based (Recommended for Most Agencies)
List each deliverable with a note that AI was involved. Price reflects the value of the output.
Campaign brief (AI-assisted) £1,200
Blog posts x 6 (AI-generated, human-edited) £2,400
Social media pack — 30 posts (AI-generated) £1,800
Ad copy variants x 12 (AI-generated) £960
Human creative direction and oversight £2,800
---
Total £9,160This approach is client-friendly. Clients see what they got. The AI notation provides transparency. Pricing reflects output value, not production method.
Approach 2: Split AI and Human Costs
Separate AI production from human oversight. Works for clients who specifically want cost transparency.
AI content generation — 6 blog posts £480
Human editing and brand review — 6 posts £1,920
AI ad copy generation — 12 variants £190
Human creative review — 12 variants £770
---
Total £3,360This approach shows clients exactly where their money goes. The risk: clients may question why human review costs so much more than AI generation and push for lower fees.
Approach 3: Bundled with Disclosure
Single line items with no AI/human cost split, but disclosure in the engagement agreement that AI agents are used.
Blog content — 6 posts £2,400
Social media management — 30 posts £1,800
Ad copy — 12 variants £960
---
Total £5,160Simple and clean. Works when the engagement letter states that AI agents assist in content production and the client has agreed to this approach. The disclosure happens at contract level, not invoice level.
Approach 4: Outcome-Based
No mention of production method. Pricing tied to results.
Lead generation content package — Q2 £8,500
Includes: 6 blog posts, 30 social posts,
12 ad variants, campaign brief
Performance target: 500 qualified leadsThis approach works for clients who care about results, not process. It aligns well with value-based pricing models that decouple fees from production methods.
How Do You Separate AI Cost from Human Review Cost?
Accurate cost separation requires tracking at the task level.
Track AI agent cost per deliverable. Every time an AI agent produces a draft, calculate the cost: tokens consumed, inference compute, tool calls, orchestration overhead. Attribute that cost to the specific client deliverable. For a blog post, the AI generation cost might be £15–40 depending on length and complexity.
Track human time per deliverable. When a human reviews, edits, or refines an AI-generated deliverable, log that time against the same deliverable. For a blog post, human editing typically takes 30–60 minutes.
Calculate the blended cost. Add AI cost + human cost + overhead margin. This is your true cost per deliverable. Your price to the client should exceed this number by enough to support the business.
Decide what to show the client. You now have the data to present costs at any level of detail. Most agencies find that deliverable-based pricing (Approach 1 above) works best. Show the AI notation for transparency. Price on value. Keep the detailed cost data for internal margin analysis.
Avoid discounting AI work to zero. Some agencies, pressured by clients, price AI-generated work at cost or below. This is a margin trap. The AI output has value. The infrastructure to produce it has cost. The human oversight that makes it client-ready has cost. Price accordingly. For guidance on fair pricing for AI-assisted work, see our pricing guide.
How Do You Handle Revisions and Iterations?
Revisions are where AI invoicing gets messy. Clear policies prevent disputes.
Client-requested revisions vs quality-driven revisions. If the client asks for changes to an AI-generated deliverable, that is a client revision. It should be billable under the same terms as any other revision. If the agency identifies errors in the AI output during review, that is a quality-driven revision. It should not be charged separately — it is part of the production process.
Set revision limits in scope agreements. Standard practice: include 1-2 rounds of revisions in the deliverable price. Additional rounds at an agreed hourly or per-revision rate. State this clearly in the proposal. Do not wait for the third round of revisions to raise the topic.
Re-running an AI agent: new charge or included? If a client’s revision requires re-prompting the AI agent — “Change the tone from formal to conversational across all 6 posts” — the agency incurs new AI costs. Whether to charge depends on the scope agreement. If the original brief specified formal tone and the client changes their mind, that is a new charge. If the agency misunderstood the brief, it is not.
Version control matters. Track which version the AI produced, which version the human edited, and which version the client approved. When disputes arise — “This is not what I asked for” — version history resolves them.
Pricing models for iteration-heavy deliverables. Some deliverables — ad copy variants, design explorations, A/B test content — are inherently iterative. For these, consider per-iteration pricing or a package price that includes a specified number of AI-generated variations.
How Do You Manage Client Expectations Around AI Invoicing?
Expectation management starts at the proposal stage, not the invoice stage.
Set expectations in the proposal. State clearly: “Our team uses AI agents to accelerate production of content, research, and creative assets. All AI-generated work is reviewed and refined by experienced human professionals before delivery.” This sets the baseline. No surprises later.
Include AI terms in client agreements. Add a clause covering: AI usage in deliverable production, the firm’s quality assurance process, how AI costs appear on invoices, revision policies for AI-generated work, and data handling practices. Get this signed before work begins.
Educate without overwhelming. Clients do not need a lecture on token economics. They need to understand three things: AI helps the agency deliver faster, humans ensure quality, and the pricing reflects the value of the output. Keep the education focused on those three points.
Handle the “why am I paying for machine work?” conversation. This question comes from a misunderstanding. The client is not paying for the machine. They are paying for the expertise to direct the machine, the judgement to evaluate its output, the skills to refine it to client standards, and the infrastructure to run it reliably. Frame the answer around what the client receives, not how it was produced.
Build trust through consistency. Use the same invoice format every month. Report AI involvement the same way every time. When clients know what to expect, they stop questioning the process and focus on the outcomes. For more on building trust through transparent billing practices, see our dedicated guide.
What Should Agencies Stop Doing?
Five common mistakes to avoid.
Hiding AI usage. The worst approach. When clients discover it — and they will — the relationship is damaged. A 2025 study found that client trust scores dropped 42% when AI usage was discovered rather than disclosed.
Charging human hourly rates for AI agent time. If an AI agent produces a blog post draft in 8 minutes and the agency bills 4 hours at £150/hour, that is indefensible. Price on value or disclose the production method. Do not charge time that was not spent.
Treating AI cost as pure margin. AI work has real costs. Treating the gap between AI cost and human-equivalent cost as pure profit works until a competitor offers the same service at a lower price. Build margin, but build it on value, not opacity.
Inconsistent disclosure across clients. If one client gets full AI transparency and another gets none, the agency is one referral conversation away from a trust crisis. Standardise the approach.
Ignoring the invoice when the engagement letter covers it. The engagement letter says AI is used. The invoice shows standard line items with no AI notation. The client reads the invoice, not the engagement letter. Put the disclosure where the client looks.
Frequently Asked Questions
How do agencies invoice for AI-generated deliverables?
Most agencies use deliverable-based line items with an AI notation — for example, “Blog posts x 6 (AI-generated, human-edited) — £2,400.” Pricing reflects the value of the output. The AI label provides transparency. Some agencies split AI and human costs on separate lines for clients who want detailed visibility.
What counts as an AI-generated deliverable?
Any deliverable where an AI agent produced the initial output: content drafts, design concepts, code, research reports. The key test: if removing the AI contribution would require the human to start from scratch, it is AI-generated. If the human work stands alone and AI enhanced it, it is AI-assisted.
Should agencies separate AI and human costs on invoices?
It depends on the client relationship. Some clients want full cost transparency. Others prefer bundled deliverable pricing. A good default: deliverable-based pricing with an AI notation, with detailed cost breakdowns available on request.
How do you handle revisions for AI-generated work?
Include 1-2 revision rounds in the deliverable price. Additional revisions at an agreed rate. Distinguish between client-requested revisions (billable) and quality-driven corrections (included in production cost). If revisions require re-running the AI agent, charge for new AI costs when the revision stems from a client scope change.
Do clients expect to pay less for AI-generated deliverables?
Some do. But the 2025 data shows that 68% of professional services buyers accept AI-assisted pricing when disclosed upfront. Clients who understand AI economics focus on value. Frame the conversation around what they receive, not how it was produced.
Keito tracks what your AI agents produce, how long it takes, and what it costs — so your invoices tell the full story. Start tracking AI deliverables today.