AD Search Studio

Content Strategy

Using AI for SEO Content Without Losing E-E-A-T: My Workflow

Key takeaways

  • Google judges content on whether it helps people, not on whether AI was involved. Mass-produced pages made to game rankings count as spam.
  • Use AI for structure and first drafts. Use people for experience, original data, judgement and voice.
  • Get subject-matter expert input before the draft exists, not after, so the draft is built around it.
  • No piece goes live without passing a pre-publish checklist covering facts, first-hand detail, voice and intent.

You can use AI for SEO content without losing E-E-A-T if AI handles structure and first drafts while people supply what a model cannot: first-hand experience, original data, expert judgement and a named author who stands behind the piece. Google's guidance focuses on whether content is helpful and made for people, not on how it was produced. What it treats as spam is content mass-produced mainly to manipulate rankings.

At Garage Collective I lead an 8-person content, technical SEO and off-page team across a 12+ brand portfolio. LLM-drafted outlines and first drafts, refined with SME input, original data and practitioner detail, are what let us publish 10 to 12 long-form pieces a month without the work turning generic. This post is that workflow: the stages, who owns each one, and the checklist that decides whether a piece goes live.

What Google actually says about AI content

It is worth getting this right, because most of the bad advice on AI content comes from misreading it.

Google's position, paraphrased from its Search Central guidance: it rewards helpful, reliable, people-first content regardless of whether it was written by a person, with AI assistance, or both. Using automation is not against the rules in itself. Using it, or anything else, mainly to manipulate search rankings is.

Since 2024, its spam policies have a name for the second case: scaled content abuse. That covers producing large numbers of pages mainly to rank, with little value for the reader. The policy applies however the pages are made, so a room of people churning out keyword variations is as much a problem as a script doing it.

Google's helpful content guidance also encourages you to think about "who, how and why": who created the content, how it was created, and why. The "why" is the one that matters most. If the answer is "to get traffic" rather than "to help the person searching", the production method is irrelevant.

E-E-A-T (experience, expertise, authoritativeness and trust) comes from Google's search quality rater guidelines. Raters do not set rankings directly, but the guidelines describe what Google's systems try to reward. The first E, experience, is where unedited AI content falls apart. A model has never run a campaign, used a product or sat in a sales call.

Where AI helps and where it does not

In my experience, the split is clear once you have done it for a while.

Good use of AIKeep it with people
Turning a brief and SERP notes into an outlineDeciding what the piece should argue
Listing questions a section should answerAnswering them from real experience
A first draft built from SME notesNumbers, examples and claims that need to be true
Summarising call transcripts and support ticketsChoosing which insight is worth publishing
Title, meta and FAQ variations to choose fromFinal voice, tone and opinion
Spotting gaps against a competitor outlineJudging whether the gap is worth filling

The pattern: AI is good at structure and at turning material you already have into readable prose. It is bad at producing the material. If you ask it for facts, examples or statistics, it will give you plausible ones, and plausible is not the same as true.

The workflow, stage by stage

This is the order we run it in. The most important choice is that SME input comes before the first draft. If you bolt expertise on at the end, you get a generic article with a few expert quotes stapled to it.

StageWhoWhat happensOutput
1. BriefSEO strategistTarget query, intent, funnel stage, what ranks now, what we know that those pages do not, internal links, named SMEOne-page brief
2. OutlineAI, checked by writerModel drafts an outline from the brief and SERP notes, with open questions marked for the SMEOutline with gaps flagged
3. SME inputSME, run by writerShort call or voice note answering the flagged questions. Real examples, data, opinions, mistakes they have seenTranscript and notes
4. First draftAIModel writes from the outline plus SME notes only. Missing facts are left as marked placeholders, never filled inRough draft
5. RewriteWriterRebuilds the draft in the brand voice, adds original data, fills placeholders from sources, cuts fillerWriter's draft
6. Expert reviewSMEChecks accuracy and adds anything that is missing or wrongApproved facts
7. Edit and publishEditor, then SEOVoice, structure, checklist, on-page SEO, schema, internal linksLive page

Stage 1: the brief is where most quality is decided

A weak brief produces a weak draft no matter who or what writes it. The line I care most about is "what do we know that the ranking pages do not". For a B2B software client like MoneyGence, that might be how accountants actually use a feature. For a D2C brand, it might be what customers ask before buying. If nobody can fill in that line, the piece is not ready to be written.

Stage 2 and 4: how we prompt

Two instructions make the biggest difference to AI drafts. First, give the model the brief, the SME transcript and your notes, and tell it to use only that material. Second, tell it to leave a visible placeholder such as [SME: example needed] or [SOURCE NEEDED] wherever it lacks information. A draft full of placeholders is useful. A draft where the model quietly made things up is dangerous, because it reads fine.

Stage 3: getting SME time without wasting it

Experts are busy, so do not send them a blank page. Send the outline with specific questions. "What do people get wrong about this?", "What would you do first?" and "Can you give one real example?" get better material than "Any thoughts?". In my experience, a recorded conversation, transcribed and summarised by AI, is the most efficient format.

Stage 5: original data and practitioner detail

This is where E-E-A-T is actually added. Sources of original material that most companies already have:

  • Search Console and analytics data from your own site
  • Questions from support tickets, sales calls and reviews
  • Product usage patterns, with permission and without identifying anyone
  • Your own tests, screenshots and before and after examples
  • Things your team has tried that did not work

One specific detail from real work does more for a page than another 500 words of general explanation.

Stage 7: brand voice

Give each brand a short style sheet: tone, words to use, words to avoid, how to address the reader, and two or three example paragraphs that sound right. The editor checks every draft against it. AI drafts drift towards the same flat, slightly formal voice, so the edit is where a Kaayu Rituals piece and a Utkarsh Small Finance Bank piece stop sounding alike.

Free checklist

The AI Citation Checklist

39 checks to get your brand cited in AI Overviews, ChatGPT, Perplexity and Gemini

  • 39 tick-box checks in six sections, in the order we do the work
  • Which AI crawlers to allow in robots.txt, and what each one controls
  • Answer-first formatting rules for passages AI tools can quote

Free. Instant access. No spam, unsubscribe any time.

The pre-publish quality checklist

Every piece passes this before it goes live. If one item fails, it goes back a stage.

  • The opening answers the target query directly in two or three sentences.
  • Every statistic, claim and quote is traced to a source we can link to or to our own data. Anything we cannot verify is removed.
  • There is at least one piece of first-hand detail: an example, a number from our own work, a screenshot or a mistake we made.
  • The SME has signed off on accuracy.
  • No placeholder text, no invented examples, no "studies show" without a study.
  • It reads in the brand voice. The editor has removed stock AI phrasing and filler.
  • The page matches the search intent (format, depth and angle) of what currently ranks, and adds something those pages do not have.
  • The named author is a real person with relevant experience, with an author bio that shows it.
  • Title, meta description, headings and internal links are in place, including links from older related pages.
  • Structured data is valid and matches what is visible on the page.
  • Product details, prices and steps are current as of the publish date.

The answer-first formatting and clean structure also make pieces easier for AI search to cite, which I go into in how to get cited in Google AI Overviews.

Signs your AI content is drifting towards scaled content abuse

Nobody sets out to break the spam policy. Teams drift into it when volume targets outrun review. Watch for these:

  • Pages that exist for each keyword variation, city or product combination with only the nouns swapped.
  • Publishing faster than anyone with expertise can read the pieces.
  • Posts with no named author, or an author who could not have written about the topic.
  • Topics chosen only for search volume, with no link to what the business knows or sells.
  • Drafts going live with minimal edits because "it reads fine".

If two or more of these sound familiar, cut volume and fix the review step before publishing anything else.

How to tell if the workflow is working

Judge AI-assisted pieces the same way you judge any content. In Search Console, check that new pages get indexed and start earning impressions for their target queries. In GA4, look at engagement and key events from organic landing pages. For SaaS, connect content to MQLs, SQLs and pipeline, as I describe in my post on B2B SaaS SEO that drives pipeline.

Then keep the older pieces working. Volume creates a large library, and a large library decays. A quarterly refresh cycle is the other half of this workflow: it is where we update facts, add new practitioner detail and catch any page that has slipped.

AI did not change what good content is. It changed how fast you can produce the parts that were never the hard part. The experience, the data and the judgement still have to come from people, and that is the part worth protecting.

Abhishek Dwivedi
Written by Abhishek Dwivedi

Founder of AD Search Studio and a Senior SEO Lead with 7 years in search, running SEO and AEO/GEO programmes across D2C, B2B SaaS and banking brands.

View Abhishek's portfolio

FAQ

Commonquestions.

Still unsure? Ask us directly.

Get in touch

Google does not penalise content for being AI-assisted. Its guidance rewards helpful, people-first content however it is produced. Content mass-produced mainly to manipulate rankings, whether by AI, people or both, falls under its scaled content abuse spam policy.

Free AI search audit

See how Googleand AI see you.

Send your site and our founder reviews it personally. You get a clear, prioritised read on what is holding your search growth back.

  • Technical health: crawl, indexation and speed issues that cap growth
  • AI visibility: how ChatGPT, Perplexity and AI Overviews describe your brand
  • Content gaps: the high-intent queries competitors win and you don't
  • The three fixes worth the most revenue, in order

Free, no obligation. Your details are only used to reply to you.

Talk to the founder

Ready to bethe answer?

Tell us what you want to rank for and where growth has stalled. We will tell you straight whether search is the right lever, and what it is worth.

Abhishek DwivediFounder and Lead Strategist, AD Search Studio

Abhishek Dwivedi