How to Write SEO Blog Posts With AI in 2026: The Complete DIY Guide

A practical system for finding what people are actually searching for, turning that into useful content, and using ChatGPT, Claude or Cursor without filling the internet with more rubbish.

Ask ChatGPT to “write an SEO article about home renovations” and you can have 2,000 words before your coffee is ready. The problem is that another 10,000 businesses can generate almost exactly the same article.

Using generative AI is not the issue. Google says AI can help with research and structuring original material. The problem is publishing lots of pages that add nothing. Google now asks businesses to create non commodity content: original expertise, experience, evidence or perspective that almost anyone with the same prompt could not have produced.

So the opportunity is not:

It is:

This is the demand-first workflow we run at Made By Beings. You can do it yourself with a spreadsheet, Search Console, and one capable AI tool.

The basic idea

Do not start with a blog topic. Start with evidence of demand, then decide whether you have something worth adding.

1. What people are searching for.
2. What problem sits behind those searches.
3. Which questions are close to a purchase or enquiry.
4. Where Google already thinks your site is relevant.
5. What competitors currently provide, and what they miss.
6. What your business can contribute that an AI model cannot invent.

Only then should AI help you write. The workflow looks like this:

AI can help at almost every stage. It should not skip them. If you only remember one rule: use AI to remove the boring work, not the thinking.

Research demand

Step 1: Start with the searches Google is already showing you

If the site already gets organic impressions, start in Google Search Console. This beats a keyword tool guessing at volume, because these are searches where Google already thinks you might be relevant.

Open Search Console → Performance → Search results. Set the date range to the last three to six months. If the branded-queries filter is available, use it so you can look at non-branded searches separately. Those are the growth opportunity.

Use Query Groups to cluster similar searches into one subject instead of analysing dozens of near-identical phrases. See Google’s Query Groups announcement.

What to export

Export Queries and Pages with impressions, clicks, click-through rate and average position. Drop them into a spreadsheet and highlight three buckets:

1. Positions 5–20. Easier to improve an existing page than to invent a new one. Google is already saying this page might answer the question, but better options exist.
2. Impressions with almost no clicks. The title may not match intent, the page may only half-answer the question, or Google may be testing you. Search the query yourself before rewriting the title.
3. Query Groups that are growing. Emerging customer interests before they become an obvious content-calendar topic.

Step 2: Find demand outside your website

Search Console only shows searches that already produced an impression. Spend 20 minutes collecting language you are currently missing. Put every phrase into the same spreadsheet, with a column for source.

Google autocomplete

Type your product, service or problem into Google, then add modifiers: best, vs, cost, price, how, why, worth it, near me, for beginners, for families, Australia. This is not volume data. It is evidence that people use that language.

People Also Ask

Search the main topic and copy the related questions. Pay attention to the shape of the question. what are south sea pearls is early research. are south sea pearls worth the price is closer to a purchase. Those should not be treated as the same page.

Related searches and Trends

Scroll to related searches and collect useful variations. Do not create one URL per phrase. Google can understand related meanings without exact matching. We are mapping a topic, not manufacturing hundreds of pages.

In Google Trends, compare two or three major concepts. You want direction, not precision. A smaller term that is accelerating can beat a larger term that is going nowhere.

Step 3: Mine the questions customers already ask

Some of the best topics never show up in keyword tools. Dump the last few months of sales emails, support tickets, live chat, reviews, call notes, social comments, Reddit, industry forums, and on-site search into the same list.

If your site has search, GA4 records the search_term parameter. Those are customers handing you content ideas. If 40 people searched afterpay and you barely mention payment options, write that page. If sales constantly hear “Can I do this with an existing house?”, that should live on the website too.

Step 4: Check what Google is finding through social

Search Console can now create Platform Properties for Instagram, TikTok, X and YouTube, so you can see how posts appear in Search, Discover and Google News.

A builder posts a Reel about a cracked slab. Six months later that Reel gets impressions for why does concrete crack, cracks in new concrete slab and is concrete cracking normal — but there is still no useful article on the website. That Reel did the market research. Write the article, then use it to support the next round of video.

Analyse opportunities

Step 5: Ask AI to build an intent map, not invent keywords

Now paste the spreadsheet into ChatGPT, Claude or Cursor. Do not ask for “100 high volume low competition keywords”. Unless the model is connected to live keyword data, it does not know volume or difficulty. It will invent both.

Ask it to group observed queries by customer intent instead:

I am researching organic search opportunities for [BUSINESS].

Below are real search queries and customer questions collected from Search Console, Google, customer enquiries, reviews and other sources.

Analyse them without inventing search volume.

Group them according to the underlying customer intent.

For each group identify:

1. The core question or problem.
2. The likely stage of the customer journey.
3. Commercial relevance to the business.
4. Whether the intent should be addressed by an existing service or product page, a new landing page, a guide, an FAQ, a comparison, a tool, or another content format.
5. Closely related questions that should probably be answered together rather than as separate pages.
6. Important questions that appear to be missing from our dataset.

Clearly distinguish observed queries from questions you have inferred.

That last line matters. AI is excellent at finding patterns. It is not a keyword database. Treat inferred questions as hypotheses until you see them in Search Console, autocomplete, or a real customer conversation.

Step 6: Think in query fan out

Google has documented query fan out: for complicated questions, its AI systems may run several related searches at once before answering.

Someone searching “Where should we stay in the Kimberley with kids if we want something more interesting than a normal resort?” is not asking one keyword. They need family accommodation, things to do with kids, options outside Broome, cultural experiences, driving distances, best time to visit, tours, food, and how many nights to stay.

That is much more useful than obsessing over a single phrase. Ask your AI to break the primary question into genuine information needs:

Take the primary customer question below.

Break it into the secondary questions someone would reasonably need answered before they could make a confident decision.

Do not create variations simply because wording can change.

Identify genuinely different informational needs.

Separate them into:

Discovery questions
Understanding questions
Comparison questions
Objection questions
Decision questions
Purchase or enquiry questions

You now have a topic map. Answer related needs on one strong page instead of spinning up a new URL for every wording.

Step 7: Check the live search results

This step stops a lot of AI rubbish. Search the primary question yourself and open the strongest results. Note the page type Google prefers, the questions everyone answers, and the gaps: generic copy, outdated advice, missing firsthand experience.

If Google shows ecommerce categories, a 4,000-word blog post is probably the wrong format. If it shows guides, a product category is the wrong format. Search intent beats your content calendar.

Step 8: Score the opportunities

You will have more topics than you can produce. Score each one from 0–5 on demand evidence, commercial relevance, existing authority, competitive weakness, and originality potential. A smaller question asked immediately before someone buys often beats a huge informational query you cannot uniquely answer.

Copy this worksheet for each topic:

Topic:
Primary question:

| Criterion | Score (0-5) | Notes |
|-----------|-------------|-------|
| Demand evidence | | |
| Commercial relevance | | |
| Existing authority | | |
| Competitive weakness | | |
| Originality potential | | |

Total:
Decision (write now / brief first / park):

Write the highest-scoring topics first. Park anything that scores poorly on originality until you have a real contribution.

Produce the content

Step 9: Decide why this page deserves to exist

This is the most important step. Before anyone writes, put a heading in the brief: Why does this deserve to exist? You need a real answer: internal data, an experiment, firsthand experience, an expert interview, customer quotes, original photos, pricing, mistakes you have actually made, a calculator, a process you use, or a comparison of products you have tested.

Google specifically wants unique viewpoints and firsthand experience, not a summary of what already exists. The anti-slop test is simple: if ChatGPT could have written essentially the same article without knowing your business, you have not added enough yet.

Step 10: Build a source pack before you draft

Give the model evidence, not vibes. Put official docs, studies, internal notes, interview quotes, product data, screenshots and competitor pages (for comparison, not copying) in one folder. Then label every claim:

Verified external facts:
Internal company facts:
Expert opinion:
Inference:
Unverified claims (do not publish without review):

Files / links:

Tell the AI to keep verified facts, internal knowledge, expert opinion and inference separate. That is how you stop confident nonsense.

Step 11: Write the brief before the article

Approve the brief and outline before anyone drafts. Copy this into your notes or AI project:

Primary search intent:
Audience:
Article promise:
Primary topic:
Supporting questions:
Original contribution:
Evidence required:
Commercial connection:
Desired next action:

Outline:
1.
2.
3.

Step 12: Install a reusable AI content system

Do not paste a giant prompt every time. Turn the process into reusable instructions. ChatGPT, Claude and Cursor can all run the same workflow if you set them up once.

ChatGPT

Create a Project for SEO content. Add brand guidelines, strong existing articles, product information, customer research and this workflow. Paste the Skill below into Project settings → Instructions.

Claude Code

Save a SKILL.md in .claude/skills/seo_content/SKILL.md. Claude can load it automatically when you are researching or drafting search content.

Cursor

The same file can live in .cursor/skills/seo_content/SKILL.md or .agents/skills/. Cursor can also pick up compatible Skills from Claude directories, so you do not need a separate prompt library.

Reusable SEO Content Skill

For ChatGPT, copy the instructions below into your Project Instructions.

For Claude or Cursor, save them as a SKILL.md file.

---
name: seo_content
description: Researches, briefs, drafts and reviews helpful search content using real demand evidence, verified sources, business expertise and anti slop quality controls. Use whenever researching or creating organic search content.
---

# SEO Content System

## Principle

SEO is demand research first and content production second.

Never invent keyword metrics.

Never manufacture expertise.

Never create pages solely to capture minor wording variations.

Use AI to analyse evidence, organise research, challenge assumptions, structure content and accelerate drafting.

The final content must provide genuine value beyond what a generic model could produce without access to the business.

## Stage 1: Understand the business

Before recommending topics, establish:

1. What the business sells.
2. Who buys it.
3. Geographic markets.
4. Most valuable products or services.
5. Customer objections.
6. Genuine areas of expertise.
7. Available proprietary data, experience, examples or experts.
8. Existing website content.

If this information has already been provided in project files, use it instead of asking again.

## Stage 2: Gather demand evidence

Prefer observed data such as:

1. Google Search Console queries.
2. Search Console Query Groups.
3. Non branded searches.
4. Google autocomplete.
5. People Also Ask.
6. Related searches.
7. Google Trends.
8. Website search terms.
9. Customer enquiries.
10. Reviews.
11. Sales questions.
12. Reddit and industry discussions.
13. Social and video search performance.
14. Existing organic rankings.

Clearly label AI generated ideas as inferred rather than observed demand.

Never claim an inferred query has search volume unless external data proves it.

## Stage 3: Cluster by intent

Cluster different queries according to genuine customer needs rather than minor wording changes.

Identify:

1. Discovery intent.
2. Informational intent.
3. Comparison intent.
4. Commercial investigation.
5. Transactional intent.
6. Support intent.

Recommend whether each need belongs on:

1. An existing page.
2. A service page.
3. A product or category page.
4. A new article.
5. A comparison.
6. An FAQ.
7. A tool.
8. A video.
9. Another format.

Avoid unnecessary content cannibalisation.

## Stage 4: Evaluate opportunities

Score each opportunity for:

1. Evidence of demand.
2. Commercial relevance.
3. Business authority.
4. Competitive weakness.
5. Originality potential.

Explain the reasoning behind each score.

## Stage 5: Analyse the search results

Before drafting, research the live search landscape when browsing is available.

Identify:

1. Dominant search intent.
2. Common content format.
3. Questions consistently answered.
4. Missing information.
5. Weak or outdated information.
6. Evidence Google values firsthand discussion, video, ecommerce, local results or another content type.
7. Opportunities to create something materially more useful.

Never copy a competitor outline simply because it ranks.

## Stage 6: Require an original contribution

Before drafting, answer:

Why does this page deserve to exist?

Identify at least one genuine differentiator such as:

1. Firsthand expertise.
2. Proprietary data.
3. Original examples.
4. Expert commentary.
5. Customer evidence.
6. Original research.
7. Testing.
8. Strong supported opinion.
9. Original photography or video.
10. A useful framework, template, calculator or tool.

If no original contribution exists, flag this before writing.

## Stage 7: Build a source pack

Prioritise primary and authoritative sources.

Separate:

1. Verified external facts.
2. Internal company facts.
3. Expert opinion.
4. Inference.
5. Unverified claims.

Never fabricate citations, statistics, customers, quotes, experience, products or outcomes.

## Stage 8: Produce the brief

Create:

1. Search intent.
2. Audience.
3. Article promise.
4. Primary topic.
5. Supporting questions.
6. Original contribution.
7. Required evidence.
8. Recommended structure.
9. Internal linking opportunities.
10. Commercial next action.

Do not draft until the brief is coherent.

## Stage 9: Draft

Write for the reader first.

Answer the core question early.

Use headings because they help humans navigate the page, not because every heading needs a keyword.

Prefer concrete language.

Prefer examples to abstractions.

Prefer useful specificity to unnecessary length.

Do not target a predetermined word count.

Do not repeat a point simply to extend the article.

Do not insert the target phrase unnaturally.

Do not pretend the company has experience it has not demonstrated.

## Stage 10: Anti slop review

After drafting, conduct a separate editorial pass.

Remove:

1. Generic introductions.
2. Obvious statements.
3. Repeated conclusions.
4. Empty transitions.
5. Corporate filler.
6. Unsupported superlatives.
7. Fake certainty.
8. Keyword repetition.
9. Formulaic AI phrasing.
10. Sections that merely restate common knowledge.

For every major section ask:

Could a generic AI have written this without our research?

If yes, either improve it with genuine information or remove it.

## Stage 11: Helpful content review

Check whether:

1. The article fully satisfies the likely reader goal.
2. The reader would need another search to understand something important.
3. Claims are supported.
4. Experience is represented accurately.
5. The content fits the business's genuine expertise.
6. The page is meaningfully different from competing results.
7. The title accurately describes the content.
8. The article avoids unnecessary length.
9. The article has a logical next action.

## Stage 12: SEO review

Review:

1. Page title.
2. H1.
3. Search intent alignment.
4. Intro.
5. Heading hierarchy.
6. Internal links.
7. Relevant external citations.
8. Image opportunities.
9. Video opportunities.
10. Metadata.
11. Structured data only where genuinely applicable.
12. Indexability and canonical setup when technical access exists.

Do not recommend AI specific markup purely for generative search.

## Stage 13: Final adversarial review

Act as a skeptical editor.

Identify:

1. Anything generic.
2. Anything unverifiable.
3. Anything written primarily for a search engine.
4. Anything missing from the reader's decision process.
5. Any stronger counterargument.
6. Any place where firsthand evidence would improve credibility.
7. Any section that should simply be deleted.

Revise before considering the article complete.

Publish and measure

Step 13: Do not “humanise” bad content

Tools that write generic AI copy and then run another model to make it “sound human” miss the point. Detectors are not the problem. Nothing valuable happened before the writing. You cannot humanise fake expertise or polish an article into having original data. Fix the research, then draft.

Step 14: Run the anti-slop test

Before publishing, highlight every paragraph that is obvious, could apply to any company, repeats something already said, makes an unsupported claim, sounds like marketing, or exists only to make the article longer. Delete those. Google has no preferred word count. The test is whether the reader can do the thing they came to do.

Step 15: Optimise without ruining the article

Once the article is actually useful, do the boring SEO: title and H1 match the searcher’s job, intro gets to the point, supporting questions are answered, internal pages are linked, claims have sources, images have alt text, the page is indexable, and structured data is used only where it applies.

There is no special “AI schema”. Google says llms.txt and rewriting pages for AI systems are unnecessary. For technical performance that does affect rankings, see mobile website performance in 2026.

Step 16: Measure, then add something useful

Publishing starts the next research cycle. Watch Search Console for new queries, growing groups, impressions climbing, positions 5–20, weak click-through rates, and questions you did not plan for. Update the page when you have something real to add. Do not change the date or shuffle copy to fake freshness. Google warns against that.

An advanced source: AI grounding queries

Bing Webmaster Tools has an AI Performance report showing cited pages and the grounding queries used in AI answers. That is how a retrieval system broke the subject down before deciding your page was relevant. Feed those phrases back into topic research. See Bing Webmaster Tools.

TL;DR

If you want help running this workflow on your site, explore consulting or start a project with Beings. For AI tooling efficiency in your own stack, see how we cut Cursor and Claude Code token spend.