Human-in-the-Loop AI SEO Workflow
Automate repetitive SEO research while humans supply judgment and originality
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 98%
The workflow divides SEO production according to comparative advantage. AI searches and scrapes leading results, examines related questions, summarizes recurring coverage, and preserves useful benchmarks or references. A human then distills those findings, decides what deserves inclusion, and supplies the product knowledge, customer evidence, market data, or expert perspective needed for information gain. AI turns the approved inputs into a brief and initial draft, but editors review both stages and polish the final article for accuracy, brand voice, usefulness, and quality. This arrangement does not claim that AI inherently ranks better. Instead, it removes repetitive research and drafting work so experienced people can produce more content while concentrating their effort on differentiation, judgment, and creation.
Origin
Extracted from Marketing Against The Grain, where Chang Chen described the workflow used to help an AI product reach approximately three million monthly visits in six months.
Core principles
- 01Use AI for repetitive, energy-draining work
- 02Keep humans responsible for judgment and quality
- 03Ground every article in search intent and SERP evidence
- 04Add information competitors and generic models cannot supply
- 05Treat AI as a productivity multiplier rather than a replacement
How to run it
- 1
Choose the target keyword
Start with a keyword that is relevant to the product, realistic for the site's authority, and connected to a recognizable user intent.
Pro tip Classify the likely intent before deciding what kind of page to create.
Watch out A high-volume keyword may still be a poor target for a new or low-authority site.
- 2
Automate SERP research
Connect AI to search results so it can inspect leading pages, identify their major components, and collect relevant supporting information.
Pro tip Include related questions and long-tail queries from the people-also-ask section.
Watch out Do not treat scraped competitor coverage as sufficient differentiation.
- 3
Distill the evidence manually
Have a human review the research, select useful findings, and decide what the article should cover or challenge.
Pro tip Preserve credible benchmarks and their reference links for later citation.
Watch out Automated summaries can omit nuance or elevate unreliable claims.
- 4
Build and review the brief
Let AI combine the approved research and human direction into a structured article brief, then have a person verify its logic and completeness.
Pro tip Specify the intended reader, search intent, product relevance, and desired information gain.
Watch out A polished brief can still be generic if its inputs contain no original insight.
- 5
Inject unique information
Add proprietary product knowledge, market research, user discussions, reviews, editorial perspectives, or subject-matter-expert insights.
Pro tip Choose evidence that directly helps the searcher rather than inserting uniqueness for its own sake.
Watch out Rewording common SERP material does not create genuine information gain.
- 6
Draft with AI
Use AI to produce the initial article from the approved brief, research, references, and original inputs.
Pro tip Constrain the draft with explicit sourcing and brand instructions.
Watch out Do not publish the generated draft without substantive review.
- 7
Polish and quality-check
An editor or writer should improve the draft, confirm references, preserve the brand voice, and ensure it fully satisfies the intended query.
Pro tip Review the page as a searcher would and remove text that adds length without value.
Watch out Scaling publication before validating quality can multiply weak content.
In the wild
A SaaS team targets a realistic comparison query. AI reviews the leading results and related questions, while an editor identifies missing implementation guidance. The team adds anonymized customer observations and original product benchmarks, asks AI to draft from the approved brief, and then edits the article for accuracy and brand voice before publication.
→ The team publishes differentiated search content faster without reducing editorial oversight.
Chang Chen described using a mixed AI-and-human production process for an AI product. AI handled research and drafting tasks while humans reviewed the work and supplied unique content, helping the operation scale without relying on unreviewed AI-only articles.
→ The product reportedly reached approximately three million monthly visits within six months while content-creation costs fell.
Common mistakes
Publishing AI-only copy
Allowing a model to write and publish the entire article removes the human judgment and original evidence that make the workflow defensible.
Confusing length with quality
More generated text does not automatically provide information gain, satisfy intent, or improve ranking potential.
Automating before validating
Scaling an unproven workflow produces weak pages faster and makes quality problems more expensive to correct.
Is it for you?
Best for
It is best for content teams that need to scale informational articles while retaining brand expertise and human accountability.
Not ideal for
It is not ideal for organizations lacking proprietary knowledge, editorial reviewers, or a credible keyword strategy.
From the transcript
“The process we are using right now: we still start with keyword, and previously, we have human to do research and to find the top…”
“After that, we have the human review to distill the learnings and to identify good part of content that we should add, then we have…”
“So you're taking over the repetitive and the energy draining part, so the human are actually focusing on the creating part.”
From the episode
How She Scaled an AI App to 3M Visitors/Mo Using AI SEO ft Chang Chen
Chang Chen