Discussion starter: an agency question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- Agency
A client wants AI-generated content published at scale and you think it’s risky. How do you handle it without losing the account?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
I would move the conversation away from “AI or no AI”, because that is not where the risk is. Google has said plainly that using AI to create content is not against its policies; what its scaled content abuse policy targets is content produced at scale mainly to manipulate rankings, with little value for readers. So the useful questions are what the pages will contain that is not already on the web, who will check them, and how quickly you will know if they are not working. If you offer the client a version of their plan that answers those questions, most clients accept it. If they still want the risky version, get the risk and the decision in writing, and decide whether you want your name on it.
Start with what the client actually wants
A request for “AI content at scale” usually hides a goal: more organic traffic, lower content costs, coverage of a long tail competitors own, or matching a competitor who appears to be doing it. Each goal has safer routes:
Underlying goal Risky version Safer version Long-tail coverage Thousands of generated articles on loosely related keywords Templated pages built on real data the client owns, with AI assisting the writing and humans checking accuracy Lower cost per article Unedited drafts published directly AI for research, outlines and first drafts; editors and subject experts for judgement and facts Freshness Regenerating old posts with new dates AI-assisted audits that flag outdated facts for a human to update Matching a competitor Copying their volume Checking whether the competitor’s scaled pages are actually indexed and earning traffic, which often they are not Propose a pilot with agreed kill criteria
The strongest way to handle a client who is set on scale is to agree a controlled pilot before committing to the full programme. It moves the argument from opinion to evidence, which protects both of you.
- Scope: a defined set of pages, say 100, in one section of the site, with a comparable set of human-written or human-edited pages as a control where possible.
- Quality bar: written criteria for what each page must contain, such as data from the client’s own systems, a checked fact per claim, and an editor’s sign-off.
- Metrics: the share of pages indexed, impressions, clicks, engagement and conversions, plus any change in the performance of the rest of the site.
- Duration: long enough for pages to be crawled and assessed, typically a few months rather than weeks.
- Kill criteria agreed up front: for example, if fewer than half the pages are indexed after the test period, or the section’s existing pages lose visibility, the programme stops or changes.
Why this works with clients:
You are not refusing the request; you are offering a faster way to find out whether it works, with less downside. Clients who want scale are usually impatient, and a pilot with clear criteria gives them a decision within a quarter.
Put the risk in writing
Whatever the client decides, write a short risk note: what the plan is, what the risks are (poor indexing, sitewide quality signals, spam policy exposure, brand damage from errors), what safeguards you recommend, and what you will monitor. Ask the client to confirm the decision. This is not about covering yourself in a cynical sense; it is about making sure the person approving the plan understands it, and that a later traffic drop is discussed against a shared record rather than competing memories.
Other risks worth naming
- Accuracy. Generated content makes confident mistakes. In regulated sectors, a wrong claim about a product, price or legal requirement can cost more than any traffic gained.
- Brand voice. Thousands of pages in a generic voice change how the brand reads to customers.
- Regulation. In the EU, Article 50 of the AI Act, applicable from August 2026, requires disclosure when AI-generated text is published to inform the public on matters of public interest, with an exemption for content under human review and editorial responsibility. It will not apply to most product pages, but for publishers and some advice content it is another reason to keep genuine human editorial control.
- Sitewide effects. A large volume of low-value pages can weigh on how the whole site is assessed, not only the new pages.
If the client insists on the risky version
At that point you have three reasonable options. You can deliver it with the risk note signed, monitoring in place and a clear agreement on who owns the outcome. You can decline that workstream while continuing other work for the client, which is often acceptable to them if you are clear and unemotional about it. Or, if the programme would put your reputation or other clients at risk, you can step away. Agencies that deliver anything asked for tend to be blamed when it fails; agencies that set conditions tend to keep the relationship longer, even if they lose the occasional argument.
Hypothetical example:
An ecommerce client wants 5,000 AI-written buying guides in a quarter. The agency proposes 150 guides for the ten highest-value categories, each built on the client’s own sales data, return reasons and product specifications, drafted with AI and edited by a category specialist. The agreement says that if fewer than 60% are indexed and receiving impressions after three months, the plan is revisited. The pilot shows strong results for guides with genuine product insight and poor results for generic topics. The client scales the first type and drops the second, and the agency keeps the account.
Where AI genuinely helps content at scale
It is worth ending the conversation positively. AI is useful for clustering keywords, drafting outlines from real data, summarising reviews and support tickets into insights, first drafts that editors improve, updating product copy from structured specifications, and spotting outdated facts. Framing the programme around those uses shows the client you are not against AI, only against publishing at a scale nobody can check.
Reframe the risk as scale without value, not AI; offer a pilot with kill criteria; document the decision; and be willing to decline the part you cannot stand behind while keeping the rest of the relationship intact.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.