Defining the 90/10 Rule Clearly
The 90/10 rule for AI content is not a precise measurement — it is a decision framework for where value comes from in a piece of content. The principle: 90% of the value that makes content rankable and useful to a reader must come from human expertise, original research, specific examples drawn from actual experience, unique analysis, and a distinct authorial point of view. The remaining 10% — structural scaffolding, prose expansion, formatting, phrasing variations — is where AI tools earn their keep.
The reason this proportion matters is that Google's quality signals are calibrated to detect the presence or absence of genuine expertise and first-hand knowledge. A piece of content that reads fluently but contains no information a qualified person could not have generated from surface-level research is thin, regardless of word count. The 90/10 rule is a guardrail against inverting this ratio and ending up with content that is structurally complete but substantively empty.
For Indian businesses in particular, this framework is relevant because the productivity argument for AI tools is compelling and the temptation to let AI do the heavy lifting is real. A content manager at a Kochi fintech startup who needs eight blog posts per month is going to be tempted to let AI produce most of the content. The 90/10 rule gives that content manager a concrete way to evaluate where they are on the safety spectrum.
The 90%: What Only Humans Can Provide
Original Expert Insight
This is the highest-value component and the one AI tools cannot replicate: a specific perspective that comes from having worked on real problems in a specific domain. A chartered accountant in Thrissur who has filed GST returns for 200 small businesses has knowledge about the edge cases, the common errors, the GST portal quirks specific to Kerala businesses, and the practical workarounds that neither a textbook nor an AI can produce. That knowledge, in print, is worth ranking for. An AI's synthesis of public GST documentation is not.
The practical implication: before any AI tool is involved, the human expert should identify three to five specific points, examples, or experiences they want to communicate. These become the anchors around which the content is built. If a human cannot identify even three specific things to say about the topic that an AI could not produce from public sources, that is a signal that either the topic is outside the expert's domain or the content production process is inverted.
Specific Examples from Actual Work
Case studies, client outcomes, and project references — even anonymised — carry enormous credibility weight in content. "We helped a Palakkad industrial equipment supplier reduce their Google Ads cost-per-lead from ₹850 to ₹310 over five months by restructuring their keyword match types and adding negative keyword lists" is a sentence that an AI cannot write from imagination. It can only exist if a human expert inserts it. And it is exactly the kind of sentence that differentiates a genuinely authoritative post from a generic one.
Proprietary Data and Original Research
Any data that does not exist in publicly available sources gives content an information gain advantage — a term that reflects how much unique knowledge a page adds beyond what's already indexed. For an IT consultant, this might be data from client project metrics, survey results from a small sample of clients, or observations from testing a specific tool configuration that no one else has published. Original data is one of the highest-value content assets available, and it is entirely human-generated by definition.
Corrected and Verified Claims
AI tools hallucinate. They present inaccurate statistics with the same confidence as accurate ones, cite sources that do not exist, and describe processes that do not work as described. A human expert reviewing AI-generated content must verify every factual claim against primary sources. This verification process is not optional — it is a core part of the 90% human contribution. Publishing AI content with unverified claims is not just a quality issue; it creates reputational and potentially legal exposure for the business.
The 10%: Where AI Earns Its Keep
The 10% allocation for AI tools is not a limitation — it maps precisely to the content production tasks where AI genuinely saves time without compromising quality. These tasks share a common characteristic: the output quality is verifiable by the human without domain expertise, because they do not require domain knowledge to evaluate.
Structural Scaffolding
Asking an AI to "generate an outline for a blog post about GST compliance for Kerala e-commerce sellers" produces a reasonable initial structure in seconds. A human expert reviews this outline, removes sections that do not apply, adds sections that are missing, reorders for logical flow, and uses it as a writing guide. The AI's outline is a starting point, not a blueprint.
Prose Expansion
When an expert has written a rough paragraph that communicates the right idea but reads too tersely, AI can expand it into flowing prose. The expert provides: "Our experience: smaller Kerala retailers undercount their own return rates because they handle returns informally without recording them, which skews their margin calculations." The AI expands this into a readable paragraph. The insight is human; the prose is AI-assisted. This is a legitimate and efficient use.
Meta Descriptions and Title Variations
Generating ten variations of a meta description from a one-sentence brief, or generating five alternative title options for a post the human has already written, is a purely mechanical task where AI excels. The human selects the best option, possibly edits it, and moves on. No expertise is required to generate these, so the AI's structural limitations do not matter.
Formatting and Summarising
Converting a bulleted list of technical points into a coherent table, summarising a long-form post into a 150-word excerpt for social sharing, or reformatting a dense paragraph into a step-by-step numbered list — these formatting tasks are safe AI applications. The underlying content was human-generated; the AI is only changing its presentation structure.
The Four-Step Workflow in Practice
A practical content production workflow that follows the 90/10 rule looks like this:
Step 1 — The expert brief. Before AI is involved at all, the human expert writes a brief that includes: the three to five specific claims or insights they want to communicate, any data points or statistics they will cite, one or two specific client or project examples they will reference, the audience's specific problem the post addresses, and any common misconceptions in this area that the post will correct. This brief is the 90%. It cannot be delegated to AI.
Step 2 — AI drafts the structure and prose scaffolding. The AI takes the brief and generates a structural outline plus initial prose sections. These sections expand the bullets from the brief into readable paragraphs but cannot go beyond the information provided in the brief — the AI has no other source to draw from at this stage. This keeps the AI's contribution bounded.
Step 3 — The expert rewrites and enriches. The human expert reads the AI draft and rewrites the sections that are too generic, adds the specific examples and data points from the brief (which the AI rendered correctly into prose but might need real numbers inserted), corrects any errors, and rewrites the introduction and conclusion in their authentic voice. This step often adds 30–40% more content than the AI draft contained.
Step 4 — Human review for accuracy and schema. A final read for factual accuracy (checking any statistics or claims the AI might have introduced), followed by adding appropriate schema markup (Article, FAQPage if applicable), optimising the title tag and meta description, and confirming the internal links are contextually relevant. Schema is not AI's job in this workflow — it requires accurate data that only the human can verify.
Where AI Creates Risk: Three Scenarios to Avoid
Generating Content on Topics You Have No Experience With
An Alappuzha-based tour operator using AI to generate blog posts about "Cybersecurity for Indian SMEs" because it's a high-traffic topic is combining AI production with zero expertise. The AI will produce plausible-sounding content, but it will contain no verifiable first-hand knowledge. Any reader who knows the subject will immediately recognise the content as thin. Google's quality systems are calibrated to reach the same conclusion.
Publishing AI Content Without Human Review
One of the most common failure patterns in Indian digital marketing is the fully automated content pipeline: AI generates the article, an automated tool posts it to WordPress, and no human reads the output before it goes live. This removes the 90% entirely. The result is content that may rank temporarily — especially for low-competition queries — but declines as Google's quality signals accumulate negative data from user behaviour.
Misrepresenting AI as Expert Experience
The ethical and practical risk: a piece of content that presents AI-generated claims as the professional's own experience. If a cybersecurity consultant publishes an AI-generated post about "My Experience Implementing Zero Trust Architecture for Kerala Banks" but has never implemented zero trust for a bank, this misrepresents expertise. If that post ranks and clients contact them expecting that expertise, there is a professional consequences. If Google's quality systems flag the site — or a competitor reports it — there are SEO consequences too.
Frequently Asked Questions
What is the 90/10 rule for AI content and SEO?
The 90/10 rule states that 90% of the value in a well-ranked piece of content must come from human input — original expertise, specific examples, first-hand experience, proprietary data, and a distinct point of view. AI tools contribute the remaining 10%: structural scaffolding, formatting assistance, grammar polish, summarising known information, and expanding bullet points into prose. When these proportions invert — when the AI provides 90% of the substance and the human adds 10% editing — the content tends to fail both user expectations and Google's quality signals.
Can I use AI to write the full draft and then edit it?
Yes, but the editing must be substantive enough to transform the draft, not just proofread it. Reading an AI draft and fixing grammar leaves 90% of the content as AI-generated surface-level information. The editing process that produces safe, rankable content involves: replacing generic claims with specific examples from your work, correcting inaccuracies the AI produced, adding data points from primary sources you have actually read, removing sections that do not serve the reader's actual need, and rewriting in your authentic voice. If your editing takes less time than the original writing would have, you probably have not edited enough.
Where does AI genuinely help in an SEO content workflow without risk?
AI is most valuable in the low-stakes, structural parts of content production: generating title tag variations for human selection, drafting meta descriptions from a brief you provide, formatting technical content into readable tables, generating FAQ questions from a topic outline (the answers should be written by an expert), repurposing existing human-written content into social posts or email summaries, and identifying structural gaps in a draft by comparing it against competitor content outlines. These applications do not put low-quality content in front of users and do not require AI to generate novel expert knowledge.
How do I explain the 90/10 rule to a client who wants to use AI to reduce content costs?
The most useful framing for Indian business clients is a comparison to cheap construction. You can save money on materials, but if you save money on the structural foundation, the building fails. In content terms: AI can reduce the cost of formatting, structure, and surface-level expansion. It cannot reduce the cost of expertise — and expertise is what ranks. The practical question for any client is: who provides the expertise? If the answer is "the AI provides it all," the content will underperform or trigger penalties. If the answer is "our team provides the expertise and AI helps express it," that is a productive and cost-efficient workflow.