Using AI to Code or Create Content? 15 Devastating SEO Issues Nobody Warns You About

1. Confidently publishing facts that are wrong

The most dangerous AI mistake is rarely obvious nonsense. It is a statement that sounds completely believable.

What the research found

In 2025, researchers from the BBC and the European Broadcasting Union (EBU) evaluated leading AI assistants using real news stories to measure their factual accuracy:

Source: Reuters

This means ChatGPT or Claude may produce:

• Fabricated statistics
• Invented product capabilities
• Misrepresented laws or regulations
• False quotations
• Non-existent research
• Outdated guidance presented as current

This happens because a language model is designed to generate a likely and useful response. It is not inherently verifying every statement against an authoritative database before showing it to you.

Why this damages SEO

Google does not have a blanket rule banning AI-generated content.

However, Google considers it a violation when businesses use automation to create large numbers of pages primarily to manipulate rankings without adding meaningful value. This is covered by its scaled content abuse policy.

Google’s official ranking history shows multiple broad core and spam updates during 2025 and 2026, including March and May 2026 core updates and a global spam update in June 2026. Google has not described these as “AI-content penalties”, but it has had large-scale impacts across websites that have used AI for content generation at scale.

2. Creating believable but completely invented citations

An AI-generated citation can look remarkably convincing.

It may contain:

• A realistic academic title
• The name of a real researcher
• A plausible journal
• A believable publication year
• A correctly formatted reference
• A URL that looks legitimate

The paper may still not exist. Alternatively, the source may be real but fail to support the claim attached to it.

While this looks like it’s great for E-E-A-T, inventing these things will actually do more harm than good, and potentially tank your rankings.

What is E-E-A-T?

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness. It’s the framework Google uses to assess whether content is reliable and written by credible sources.

Fake citations, invented research and unsupported claims undermine trust, making it harder for both search engines and readers to believe your content.

What the research shows

A review published in Nature’s Scientific Reports analysed 732 AI-generated academic citations across six studies.

The conclusion: AI is improving, but every citation still requires human verification.

The 4 citation failures

Before publishing, always check

• The paper exists
• The authors and publication details are correct
• The source actually supports your statement
• You’re using the original research, not a secondary summary

3. Producing content that sounds exactly like everyone else

AI can make average writing more fluent. It can also make thousands of businesses sound remarkably similar.

Common patterns include:

• Identical introductory structures
• Generic lists of benefits
• Predictable headings
• Empty summaries
• Repetitive sentence rhythms
• Excessively balanced conclusions
• Overused transitions
• No original experience
• No defensible opinion
• No proprietary data
• No customer language
• No evidence that the writer has done the work

Sometimes people focus on superficial clues such as em dashes, certain adjectives or a particular sentence structure.

Those are weak indicators.

An em dash is not an SEO problem, nor does it prove that content was written by AI. Skilled human writers use the same punctuation. The real problem is not one punctuation mark. It is a lack of distinctive thinking.

The real impact of generic content on your website

Look for and create content that includes:

• Original research or data
• Real case studies
• First-hand experience
• Customer insights
• Clear opinions
• Unique processes
• Local knowledge

4. Targeting keywords without understanding search intent

AI can generate thousands of keywords in seconds, but keywords alone don’t drive rankings or revenue.

An experienced SEO doesn’t just find keywords — they understand search intent, map each keyword to the right page, analyse the live search results, and align content with what both Google and potential customers actually want.

Choosing the wrong page type can mean:

• Service pages targeting informational searches
• Blog posts targeting buying keywords
• Ignoring local intent
• Keyword cannibalisation
• Chasing traffic instead of revenue

Keyword intent

How an SEO expert validates keyword opportunities

AI can generate keyword ideas, but it can’t decide which ones will grow your business.

An experienced SEO combines keyword research with search intent, live Google results and commercial priorities to decide which pages deserve to exist.

What you can do

• Search your target keyword in Google and analyse the first page. Are the results blogs, service pages, products or local businesses?
• Ask ChatGPT to classify the search intent (informational, commercial, transactional or local), then compare it with what actually ranks.
• Check Google Search Console to see which keywords your existing pages already rank for before creating new content.
• Ask your sales team what questions customers ask before buying, then build content around those topics.
• Prioritise keywords that drive enquiries, not simply the ones with the highest search volume.

5. Publishing hundreds of pages that compete against each other

AI makes it easy to create:

• A page for every city
• A page for every service variation
• A page for every industry
• A page for every customer type
• A page for every long-tail keyword
• A blog answering every possible question

This may look like comprehensive SEO coverage.

It may instead create a website where dozens of pages say almost the same thing — causing massive cannibalisation issues as well as crawl budget issues.

Google’s scaled content abuse policy also applies regardless of whether pages are generated by AI, traditional software or human writers. The concern is producing large volumes of low-value or unoriginal pages primarily to manipulate rankings.

Check to see if your AI content has already caused you issues

• Search Google using: site:yourdomain.com "your main keyword". If you find multiple pages targeting the same keyword or answering the same question, they’re likely competing against each other. Decide whether each page has a unique purpose or whether they should be merged.
• Use Google Search Console. Go to Search Results → Queries, select an important keyword, then review the Pages tab. If several URLs receive impressions for the same query, check whether Google is unsure which page should rank.
• Review the search intent. Search the keyword in Google and analyse what already ranks. Is Google favouring service pages, blog posts, product pages or local results? Make sure your page matches the format users are expecting.
• Consolidate similar content. Instead of creating another page, strengthen your best-performing one. Merge overlapping content, redirect weaker URLs, and update internal links so Google understands which page is the primary resource.
• Create genuinely local content. For location pages, don’t just swap the city name. Include local case studies, testimonials, photos, service areas, FAQs, landmarks and examples that demonstrate real experience in that location.

6. Introducing security vulnerabilities nobody notices

AI-generated code frequently looks functional. A form submits. A login works. A search box returns results. A database stores the information.

That does not mean the implementation is secure.

Common vulnerabilities include:

• Unsanitised user inputs
• Weak authentication
• Exposed API keys
• SQL injection
• Insecure session management
• Missing security headers
• Unsafe third-party dependencies
• Sensitive data written to logs
• Credentials stored directly in source code

A simple example

An AI may generate a database query by directly inserting user input into the command:

query = "SELECT * FROM users WHERE email = '" + user_input + "'"

The feature may work in normal testing. It also creates a potential injection vulnerability.

A secure implementation would normally use parameterised queries, validate inputs and apply appropriate database permissions.

What you can do to check for security vulnerabilities

Before publishing AI-generated code, run:

• Static application security testing
• Dependency and package scanning
• Secret detection
• Authentication and permissions testing
• Input-validation tests
• Security-header checks
• Cross-site scripting tests
• SQL injection tests
• Session-management reviews
• Manual code review by an experienced developer

7. Building JavaScript-dependent websites (crawl & render issues)

The issue is not that Google cannot process JavaScript. Google can crawl, render and index many JavaScript websites.

The issue is that JavaScript adds another stage where something can fail.

When essential content, links, titles, canonical tags or navigation only appear after JavaScript executes, search engines must successfully render the page before they can understand everything.

Why this matters

A website can look perfect to visitors while Google sees something completely different.

Common AI coding mistakes include:

• Loading important content only after JavaScript executes
• Hiding internal links behind JavaScript
• Creating pages with empty HTML source
• Blocking JavaScript files needed for rendering
• Relying on client-side rendering when server-side rendering is more appropriate

What you can do

• View your page source. Use Ctrl+U or right-click → View Page Source and search for your main heading or body text. If your content isn’t present in the HTML source, search engines may need to execute JavaScript before they can see it — and there’s no guarantee they’ll render everything correctly.
• Use GSC’s URL Inspection tool. Inspect important pages. Compare the rendered page with what you see in your browser. Check whether Google can see all headings, body copy, images, navigation and internal links. If elements are missing, they may not be indexed.
• Test your website with JavaScript disabled. If the page becomes mostly blank, navigation stops working or key content disappears, your website may be overly reliant on JavaScript for SEO-critical elements.
• Use Server-Side Rendering (SSR) or Static Site Generation (SSG) for important landing pages, service pages and blog content. These approaches deliver HTML directly to search engines, making pages faster to crawl, easier to index and generally more reliable for SEO than relying entirely on client-side rendering.

8. Creating code that works today but breaks tomorrow

AI coding assistants are very good at producing a plausible solution to the immediate request.

They are less reliable at understanding:

• The organisation’s long-term architecture
• Future product requirements
• Upgrade plans
• Internal development standards
• Security policies
• Legacy integrations
• Unsupported dependencies
• What another developer will need six months later

How to avoid future issues

Before deploying AI-generated code, test far more than the feature you asked it to build.

Check different browsers, mobile devices, user journeys and edge cases to ensure the code is reliable. Ask the AI to explain how its solution works, why it chose that approach, and whether there are any limitations or better alternatives.

Keep documentation up to date so future developers understand what was changed and why. For important features, have an experienced developer review the code for security, performance and long-term maintainability.

9. Producing code nobody can maintain

It’s easy to measure how quickly AI can generate code. It’s much harder to measure the long-term cost of maintaining it.

A 2026 large-scale study analysed more than 304,000 verified AI-authored commits across 6,275 GitHub repositories, tracking code generated by five of the most widely used AI coding assistants.

Researchers identified 484,606 quality issues, with 89.1% classified as code smells — maintainability problems that don’t necessarily break software today but often increase technical debt and make future development more difficult.

• More than 15% of AI-generated commits introduced at least one detectable quality issue, regardless of which AI assistant was used.
• 24.2% of those issues were still present in the latest version of the repositories, suggesting many problems are never corrected after deployment.
• While AI sometimes removed existing code smells through refactoring, it also introduced new maintainability issues, highlighting that the outcome depends heavily on human review and quality assurance.

Read the study: arXiv

The maintainability test

Ask a developer who did not generate the code to answer:

1. What does this component do?
2. Which other systems depend on it?
3. What assumptions does it make?
4. How is it tested?
5. What happens when an external service fails?
6. Where would a new feature be added?
7. Can one part be changed without breaking another?
8. Are the names and structure consistent with the rest of the codebase?

When these questions require another long AI conversation to answer, the documentation and architecture are inadequate.

10. Fixing one problem while silently creating another

AI usually responds to the problem described in the prompt.

It does not automatically understand every dependency around that problem.

• Ask it to fix a mobile table and it may affect every table on the site.
• Ask it to improve speed and it may delay scripts required for analytics.
• Ask it to simplify navigation and it may remove crawlable internal links.
• Ask it to change a form and it may break the event used to track leads.

What you need to do

Every AI-generated change should pass through:

1. A staging environment
2. Code review
3. Automated tests
4. Mobile and desktop testing
5. Browser testing
6. Security checks
7. SEO checks
8. Accessibility checks
9. Performance checks
10. Analytics validation

GitHub recommends testing AI-generated code with static analysis, automated tests, dependency checks and collaborative review before merging it into production.

11. Destroying the website’s brand voice

AI-generated website copy often sounds polished, but it also tends to sound like everyone else.

It relies on safe, overused phrases that explain what a business does without showing any real personality, understanding or point of view.

This makes the brand harder to remember, weakens trust and gives customers fewer reasons to choose it over a competitor. Generic content can also reduce conversions because it describes the product without connecting to the customer’s real frustrations or desired outcomes.

Which one do you think will convert more sales?

How to make AI-written content more distinctive

Before asking AI to write, give it:

• The exact customer you are speaking to
• The problem they are already experiencing
• The words customers use to describe that problem
• The outcome they want
• The objections stopping them from buying
• The features that genuinely make the product different
• The tone the brand uses and avoids

Good website copy should make the customer feel understood before it asks them to buy.

12. Creating serious accessibility failures

A website can look perfectly fine to the person who built it while still being difficult, frustrating or impossible for someone else to use.

AI-generated websites often miss the context needed to make forms, navigation and interactive features genuinely accessible.

Common problems include:

• Form fields without clear labels
• Poor heading structure
• Buttons and links that are difficult to understand
• Keyboard navigation that does not work
• Low colour contrast
• Missing or unhelpful image descriptions
• Incorrect ARIA attributes
• Pop-ups that trap the user
• Errors that screen readers do not announce
• Features that only work with a mouse
• Content that breaks when zoomed

These issues can prevent people with visual, motor or cognitive impairments from completing basic actions such as reading a page, submitting a form or making a purchase.

What you can do to improve website accessibility

Start with an automated accessibility scan, but treat it as the beginning of the review, not the final result. Fix every flagged issue, then test the website manually.

Navigate each priority page using only Tab, Shift+Tab, Enter, Space and Escape. Confirm that every link, button, menu, form field and pop-up can be reached, clearly identified and operated without a mouse.

Increase the browser zoom to 200% and 400%. Check that text remains readable, content reflows correctly, buttons stay visible and no information overlaps, disappears or requires horizontal scrolling.

Review every page for clear heading order, sufficient colour contrast, descriptive link text, properly labelled forms and useful alternative text.

Submit forms incorrectly on purpose to make sure error messages explain what went wrong, identify the affected field and can be recognised by assistive technology.

Use a screen reader to complete your most valuable journeys from start to finish, including navigation, enquiries, account creation, checkout and payment. Test the full task, not just individual elements.

Use WCAG 2.2 as the minimum benchmark. Add specific accessibility requirements to the project brief, assign responsibility for testing and make accessibility approval mandatory before launch.

13. Damaging website speed and mobile usability

AI website builders can create visually impressive pages very quickly, but they often add more code, animations and media than the website actually needs. The result may look excellent on a large desktop screen while loading slowly, shifting unexpectedly or becoming difficult to use on a customer’s phone.

Common problems include bloated JavaScript, duplicate CSS, oversized images, unnecessary plugins, video backgrounds, excessive animations, multiple font files and too many third-party scripts. On mobile, this can cause text to overflow, tables to become unreadable, buttons to become difficult to tap and important content to move while the page is loading.

Website speed is not just a technical score. It directly affects whether visitors remain on the page, trust the business and complete an enquiry or purchase.

Google recommends monitoring three Core Web Vitals:

As a general target, important pages should aim for an LCP of 2.5 seconds or faster, an INP below 200 milliseconds, and minimal layout movement.

How to check and optimise your website load speed

Test your homepage, service pages, product pages and other high-value templates in PageSpeed Insights. Review both the mobile and desktop results, but prioritise mobile because this is often where the biggest problems appear.

Check Google Search Console’s Core Web Vitals report to see whether real visitors are experiencing slow loading, delayed interactions or unstable layouts.

Open the website on several real phones and test it using mobile data, not only fast office Wi-Fi. Check whether the main heading and image appear quickly, menus open correctly, forms are easy to complete, tables remain readable and buttons are large enough to tap comfortably.

Before building, set clear limits for image sizes, JavaScript, third-party scripts, fonts, animations and background videos. Compress images, remove unused code, load below-the-fold media only when needed and avoid adding plugins or tracking scripts without a clear purpose.

14. Breaking analytics and conversion tracking

An AI-generated redesign can look completely normal while quietly breaking the systems used to measure leads and sales.

The form may still submit, but:

• The conversion event no longer fires
• WhatsApp or phone clicks disappear
• Purchases are counted twice
• Campaign data is lost
• Leads reach the CRM without a source

This can make successful campaigns appear unprofitable and lead to poor marketing decisions.

The full measurement chain

Test the complete journey before publishing

1. Complete the action. Submit a real form, booking or test purchase.
2. Check the event. Confirm it fires once, at the correct moment.
3. Check the data. Verify the source, campaign, value and transaction ID.
4. Follow it through. Make sure it appears in analytics, advertising platforms and the CRM.

Also test with cookie consent accepted and rejected, and refresh confirmation pages to check that conversions are not recorded twice.

15. Creating false confidence because the output looks finished

AI output often looks complete long before it has been properly checked.

A page may be indexed, a form may submit and a citation may open, but none of those signals prove the work is accurate, secure or commercially effective.

AI removes the visible effort involved in producing work. It does not remove the hidden work required to approve it.

Expertise is knowing what to question

An experienced SEO & developer would review:

• Whether the opportunity has commercial value
• Whether the right page is targeting it
• Whether the content matches search intent
• Whether traffic is likely to generate enquiries or sales
• How the change affects the wider website
• Security and data handling
• Performance and compatibility
• Failure conditions
• Maintainability
• Long-term cost

The verification matrix

Use this matrix to separate surface-level signals from the evidence needed for genuine approval. Every important output should be tested against the outcome it is meant to achieve, not simply judged by how complete it looks.