Doing SEO With ChatGPT-What It Does Well and Where It Quietly Fails

ChatGPT is useful for parts of SEO and quietly harmful for others. The harm is quiet because ChatGPT writes a made-up figure in the same confident tone as a true one; OpenAI's own research page describes these errors as "plausible but false statements".

If you run a Shopify store, the risk shows up in your product pages, meta descriptions and keyword plans. This guide draws the line between the jobs ChatGPT can help with and the ones it can get wrong, using what OpenAI's research and Google's guidance for site owners say and what studies have measured. Then it gives you a workflow that keeps the time saved without the errors, starting with the short answer.

[S2, TL;DR]

The short answer

ChatGPT can help with SEO as a fast first-draft and sorting tool, but it cannot be the source of your facts or your search numbers. In Google's words, "appropriate use of AI or automation is not against our guidelines", and Google asks that a person fact-check all AI-written content before it goes live, including titles and meta descriptions.

• It is good at grouping keywords you already have, building outlines, drafting titles, meta descriptions and FAQ answers, writing schema code for you to test, and turning your own notes into clear copy.

• It fails at facts, figures and sources it has not been given, and it can invent them in a confident tone.

• Search volumes are a weak spot. In a two-run check on 23 October 2026, the same question got 300, then 50; Google's own data put the keyword at about 40 a month.

• Copy you let ChatGPT write alone tends to read like everyone else's, which Google calls commodity content.

• The safe workflow: start from your own data, add what only your store knows, check every claim, and publish only as fast as you can review.

The detail starts with the jobs ChatGPT does well.

[S3, Table of contents]

What ChatGPT does well for SEO

ChatGPT does well at SEO jobs where you give it the material and can check the result: sorting, structuring and first drafts. Google's own guidance says generative AI "can be particularly useful when researching a topic, and to add structure to original content".

These are the jobs where it can help most, each with the check that keeps it safe:

• Grouping keywords you already have, from Google Search Console or a keyword tool, by meaning, so each group gets one page. To check it, search the main term in Google and see what kind of page ranks.

• Outlines built from real questions, such as People Also Ask results and customer messages. Check that every heading answers a question someone asked.

• First drafts of title tags, meta descriptions and FAQ answers. Check the length, and check every product fact against your product data.

• Drafts of schema markup (structured data). Google includes structured data in the review it asks for, so run the markup through Google's Rich Results Test and make sure it matches what the page shows.

• Turning notes you write yourself into clear copy: your supplier specs, your returns policy, your delivery times. Check that nothing was added that was not in your notes.

• Small technical helpers, such as spreadsheet formulas, regex for Search Console filters and redirect lists. Test each one on a copy before you use it.

There is a measured case for the first draft too. In a study by researchers at UCL and the University of Exeter, published in Science Advances in 2024, less creative writers who were given five AI story ideas wrote stories rated up to 26.6% better written, with novelty up 10.7%. The study used short fiction, not product or SEO copy. The next sections show what that same draft gets wrong when ChatGPT is asked for facts and numbers it was not given.

[S5, Body]

Where ChatGPT quietly fails: facts and numbers

ChatGPT writes the most likely next words rather than looked-up facts, so when it does not know, it can produce an answer that sounds right and is not. OpenAI calls these hallucinations: "plausible but false statements generated by language models".

Why it guesses. OpenAI's researchers say models are graded on accuracy, so they learn to guess "rather than say 'I don't know.'" In OpenAI's own SimpleQA test, an older model, OpenAI o4-mini, declined to answer 1% of the time and gave a wrong answer 75% of the time. A newer one, gpt-5-thinking-mini, declined 52% of the time and was wrong 26% of the time. OpenAI says hallucinations "remain a fundamental challenge for all large language models".

OpenAI's SimpleQA test, as published by OpenAI

Sources it makes up. A peer-reviewed study in Scientific Reports (2023) checked 636 citations that ChatGPT was asked to write, across 84 documents on 42 topics. Of those citations, 55% from GPT-3.5 and 18% from GPT-4 did not exist. These were older models and academic sources, but the habit matters for any SEO copy that quotes a study or a statistic.

Search numbers it cannot measure. When this guide was prepared, on 23 October 2026, the same question was sent to ChatGPT twice: the average monthly Google search volume for "how to do seo" in the United Arab Emirates. It answered 300, then 50. Google's own advertising data put the keyword at about 40 a month (the average from October 2025 to September 2026). That is two runs of one question, not a study; the point is that the number changed between two asks. Take volumes from Google Keyword Planner, which shows "estimates on the number of searches a keyword gets each month", or from a keyword tool, not from ChatGPT.

Google says the same in its guidance on AI content: generative models "don't retrieve facts, but predict a likely sequence of words", and their output "may contain inaccuracies (also known as hallucinations)". There is also a quieter failure, where the copy is correct but reads like everyone else's.

[S6, Body]

Where ChatGPT quietly fails: sameness and scale

Content written by ChatGPT alone tends to say what is already on other pages, because it is built from patterns in text that already exists. Google calls this commodity content, and generating many such pages without adding value is listed in its spam rules as scaled content abuse.

The sameness has been measured. In the Science Advances study, stories written with one AI idea were about 11% (10.7%) more similar to each other than stories written without AI, and the researchers describe AI-assisted stories as "less varied and diverse". It is the same study as above: each draft gets better, and the pool of drafts gets more alike.

Google Search Central's guide for site owners gives "7 Tips for First-Time Homebuyers" as its example of commodity content, and says that creating content people find "unique, compelling, and useful will likely influence your website's presence". For example, if you let ChatGPT write a product description from the supplier's text, you get the commodity version: copy that other stores selling the same product can have too. Google's spam rules name product descriptions "copied directly from the original merchant without any original content or added value".

Scale makes it worse. Google lists "using generative AI tools or other similar tools to generate many pages without adding value for users" as scaled content abuse, and says sites that break its spam policies "may rank lower in results or not appear in results at all". Asking ChatGPT to repeat your keyword more often, an old SEO habit, does not help either. In a 2023 lab test from the Generative Engine Optimization (GEO) study, keyword stuffing lowered a page's share of the AI answer by about 8%, while adding quotations raised it by about 41%. Those figures held for pages already among the five sources the AI read, and rankings and sales were not measured. That leads to what Google says about AI-written content as a whole.

[S7, Body]

What Google says about AI-written content

Google does not ban content written with AI; it judges the content, not the tool. Google Search Central put it this way on 8 February 2023: "Appropriate use of AI or automation is not against our guidelines", and "Using AI doesn't give content any special gains. It's just content."

Google draws the line at purpose. Using automation, including AI, "with the primary purpose of manipulating ranking in search results" breaks its spam policies.

Google's current guidance, last updated on 1 October 2026, adds a clear instruction: "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing". That review covers title tags, meta descriptions, structured data and image alt text, not only the body text. Google also suggests telling readers how a piece of content was made.

One large study shows how common AI text already is. In July 2025, Ahrefs checked 600,000 top-ranking pages with its own AI detector and found that 86.5% contained some AI-written text, while 4.6% were pure AI. The link between the share of AI text and ranking position was 0.011, close to none. Ahrefs sells that detector, and detectors can be wrong, so read this as "AI use is common on pages that rank", not as proof either way.

If you let ChatGPT write product titles or descriptions and send them to Google Merchant Center, Google says they must be specified separately and labelled as AI-generated. The SEO workflow below meets these rules step by step.

[S8, Body]

How to do SEO with ChatGPT, step by step

To do SEO with ChatGPT safely, give it your data and keep the facts and the final check for yourself. Start from your own Search Console and keyword data, let ChatGPT group and draft, add facts only your store has, fact-check everything including titles and meta descriptions, and publish only as fast as you can review.

1. Start from measured data. Export the queries people already use to find you from Google Search Console, and take search volumes from Google Keyword Planner or a keyword tool. Keyword Planner sits inside a Google Ads account, and Google asks you to enter billing information to finish setting the account up. What breaks: asking ChatGPT for the numbers, which can change between two asks.

2. Group by meaning. Ask ChatGPT to group those queries by meaning and suggest one page per group. What breaks: two different needs merged into one page. Search each group's main term and check what Google shows.

3. Outline from real questions. Build each outline from People Also Ask, your customer emails and your WhatsApp messages. What breaks: questions ChatGPT makes up that no buyer asks.

4. Add what only your store knows. Before any drafting, add your prices, delivery times, what you tested, your photos and your returns policy. This is what makes SEO-optimized content worth reading. What breaks: letting ChatGPT write the whole page from the supplier's text.

5. Let ChatGPT draft. This is where it can help most: titles, meta descriptions, FAQ answers, alt text and schema. What breaks: a wrong product fact in a title, or schema that does not match the page. Test the schema in the Rich Results Test and check that the structured data matches the visible text on the page.

6. Fact-check everything. Check every number, name, source and claim before publishing, including titles, meta descriptions, structured data and alt text. This is Google's rule for AI-generated content. What breaks: a made-up statistic or study left in.

7. Publish at the pace you can review. Then watch Search Console. What breaks: publishing pages faster than anyone checks them, the pattern Google's spam rules call scaled content abuse, or sending AI-written product data to Google Merchant Center without its label.

Before you paste anything. Check your settings before you paste customer names, order data or sales reports into ChatGPT. OpenAI says that for personal ChatGPT accounts "we may use your content to train our models", unless you turn off "Improve the model for everyone" in Data controls. By default, OpenAI does not use ChatGPT Business, ChatGPT Enterprise or ChatGPT Edu content to train its models, and a Temporary Chat is not used while it stays temporary. Either way, strip personal details from exports before you paste them.

Even with this workflow, some parts of SEO still need a person.

[S9, Body]

What still needs a human

A person still has to own the parts of SEO that depend on your business and your live site: the facts, the judgment, the diagnosis, the byline and the result. ChatGPT can help with each of these, but it cannot own them.

• Facts only your business has, such as stock, prices, delivery promises and product tests. ChatGPT can only repeat what you give it.

• Judgment about which products and pages matter to your margins, and what not to publish.

• Diagnosis of the live store. ChatGPT can write a technical SEO checklist, but it is not a site crawler: it may search the web when a question needs current information, and it does not crawl your whole store. A crawler and Google Search Console show what is broken on your site.

• The byline. Google suggests you "consider having accurate author bylines when readers would reasonably expect it", and says giving AI an author byline is "probably not the best way to follow our recommendation".

• Measurement, meaning the decision on whether a change worked, made from Search Console and sales data rather than from ChatGPT's view of it.

This is also the honest test for any agency: ask who checks the facts and who answers for the result. There is also a second thing people mean when they say "ChatGPT SEO".

[S10, Body]

SEO for ChatGPT is a different job

SEO for ChatGPT means getting your pages found and cited inside ChatGPT's answers, which is a different job from using ChatGPT to do your SEO. That is what most people mean when they search how to do SEO for ChatGPT, how to optimize SEO for ChatGPT, or how to optimise SEO for AI.

Three documented facts cover the basics. ChatGPT can only show your pages in its search answers if OpenAI's search crawler, OAI-SearchBot, is allowed on your site; GPTBot is a separate crawler that collects content that may be used in training, and the two settings are independent. OpenAI also says plainly: "Placement is not guaranteed." Google's spam policies, last updated on 28 August 2026, include "attempting to manipulate generative AI responses in Google Search", so tricks sold as ways into AI answers carry the same risk as tricks for rankings.

How AI search engines choose what to cite is a separate topic from this guide, so the questions below stay with using ChatGPT for SEO.

[FAQ block, sections S11 to S15]

FAQ

[S11, FAQ]

Can ChatGPT do SEO?

ChatGPT can do parts of SEO, such as drafting, grouping keywords and writing code to test, but not the parts that need true facts, measured numbers or a look at your live site. Google treats what it writes like any other content and asks that a person fact-check it before publishing. Whether that content hurts your rankings depends on how it is used.

[S12, FAQ]

Does ChatGPT content hurt SEO?

ChatGPT content does not hurt SEO simply because it was written with AI; it hurts when it is wrong, copied in substance, or published in bulk without adding anything. Google says "using AI doesn't give content any special gains", and lists mass AI pages that add no value for users as scaled content abuse. That leaves the question of whether you can do SEO yourself.

[S13, FAQ]

Can I do SEO myself?

You can do much of SEO yourself with Google's SEO Starter Guide, Google Search Console and ChatGPT as a drafting help, as long as you keep the facts and the final check. The parts that ask the most of you are diagnosing technical problems on the live store and producing pages with information other stores do not have, and those are the parts where outside help is worth weighing. If you use ChatGPT for the drafting, how you word your prompts can lower the risk.

[S14, FAQ]

Which ChatGPT prompts work for SEO?

The ChatGPT prompts that work best for SEO limit what it may make up, rather than ask for more output. Three instructions you can copy:

• "Use only the data I paste below. If something is missing, say so."

• "If you are not sure of a fact, write 'not sure' instead of guessing." OpenAI's own research says models guess when they are not rewarded for saying "I don't know", and in its SimpleQA table the model that declined to answer more often made fewer errors.

• "List every fact and number you used, so I can check each one."

These prompts can help lower the risk; they do not remove the need to check. The same care applies when the goal moves from ranking in search to being quoted in AI answers.

[S15, FAQ]

What is the difference between SEO and AI engine optimisation?

Search engine optimization (SEO) aims to rank pages in search results, while AI engine optimisation, also called Generative Engine Optimization (GEO) or Answer Engine Optimization (AEO), aims to have a page quoted inside AI answers. Google treats work to optimise for its AI features as part of SEO. Who prepared this guide, and the sources behind it, follow below.