Google Updates Its Guidance on AI-Generated Website Content

On 1 October 2026, Google updated its Search Central guide to using generative AI content on your website. The documentation changelog describes the change as adding information from the Search Quality Rater Guidelines, to align the page with what Google has been presenting at developer events. The guide still does not ban AI-written content. It does make clearer where AI-assisted pages cross into spam, which parts of a page must be checked by a person, and what online stores must label.

For companies in Saudi Arabia, the UAE, Qatar and the wider Gulf that use AI to produce Arabic and English versions of service pages, product descriptions and blog posts, the update is a practical signal. Below is what changed, what did not, and how to adjust your workflow.

What the updated guidance says

The rater guideline references are the substantive addition. Raters are people Google uses to evaluate search results. Their scores do not move individual rankings, but the guidelines show what Google considers low quality. Sections 4.6.5 and 4.6.6 describe the two patterns AI makes easiest to produce: many near-identical pages, and pages that add little or nothing of value.

What it does not say

Two common readings of this update are wrong.

It does not say AI content is penalised for being AI. The guide treats generative AI as a legitimate tool for research and structure. The problem it describes is scale without value, which the spam policies define as “when many pages are generated for the primary purpose of manipulating search rankings”. The policy’s examples include “using generative AI tools or other similar tools to generate many pages without adding value for users” and “stitching or combining content from different web pages without adding value”. Hand-written pages produced in bulk for the same purpose fall under the same policy; what to check after the latest enforcement round is in our note on Google’s September 2026 spam update.

It does not say a rater can demote your site. The guide says plainly that ratings do not directly influence ranking. Read the rater sections as a description of what Google’s systems aim to reward, not as a list of manual penalties.

The useful question for a business, then, is not “will Google detect our AI?” but “would these pages still be worth publishing if nobody knew how they were written?”

Why this matters for business websites in the Gulf

Several habits common on Gulf company websites are exactly the patterns the guidance names. None of them is wrong in itself. Each becomes a risk when it is done at scale with no human input.

Bilingual sites generated in one pass

Most corporate sites in the region need Arabic and English. It is now cheap to write the English page with AI and generate the Arabic one from it, or the reverse. If no one who knows the business and the market reviews the result, the Arabic version often reads as a translation rather than a page written for an Arabic-speaking buyer. Wrong technical terms, a literal tone and claims that do not fit local practice are typical. That is the “little to no added value” pattern, now in two languages.

City and service pages multiplied by template

A contractor or maintenance company that serves Riyadh, Jeddah and Dammam, or Dubai, Abu Dhabi and Sharjah, may be tempted to publish one AI-written page per city per service. If the pages differ only in the city name, they match the scaled content description closely. A city page earns its place when it carries something genuinely local: the branch address and team, service areas, local approvals or projects completed there.

Distributor product catalogues

Agents and distributors often have hundreds of products and only manufacturer datasheets to work from. Generating a description for each with AI is efficient, but stitching together text from the manufacturer’s site and other resellers without adding anything is one of the listed spam examples. The added value a distributor can supply is local: availability, approvals relevant to the Saudi or UAE market, support, and how the product fits typical local projects.

Metadata nobody reads before publishing

Titles, meta descriptions, structured data and image alt text are increasingly filled in automatically by plugins and AI tools. The new guide names all four as things to fact-check, because they can appear directly in Search. A wrong service area in a meta description, or a FAQ schema answer that contradicts the page, is now explicitly your responsibility.

If you sell online: the Merchant Center rules

E-commerce brands and retailers in the region that list products in Google Shopping have firmer, more technical obligations. Google’s Merchant Center help page on AI-generated content sets three rules:

• Images: AI-generated product images must contain IPTC DigitalSourceType metadata with the value TrainedAlgorithmicMedia, and you should not strip that metadata when exporting or compressing images.
• Titles: AI-generated titles go in the structured_title attribute, with digital_source_type set to trained_algorithmic_media and the title text in content.
• Descriptions: AI-generated descriptions go in structured_description, with the same sub-attributes.

If your feed is produced by a plugin or an agency, ask directly whether any titles, descriptions or images are AI-generated and, if so, whether these attributes are populated. Image metadata is often lost silently when files are resized or run through an optimiser.

A review workflow for AI-assisted pages

The goal is not to stop using AI. It is to make sure that every page that goes live has been through a person who knows the business. A workable sequence for a marketing team:

1. Start from your own material. Give the tool your service scope, project list, approvals, branch details and real customer questions. Output built on that is far more likely to add value than output built from what the tool already knows.
2. Use AI for research and structure, then write the substance. This is the use Google describes as particularly useful. Outlines, comparisons and first drafts are reasonable; the claims, examples and local detail should come from your team.
3. Write each language for its reader. Have a native speaker who understands the sector review the Arabic and the English separately, rather than checking one against the other. Our Arabic AI copy checklist lists what that reviewer should look for.
4. Fact-check everything that can appear in Search. Title, meta description, headings, FAQ answers, structured data and alt text, as the guide lists. Validate structured data before publishing.
5. Apply a “worth publishing” test before scaling. Before generating a set of similar pages, ask what each one offers that the others do not. If the honest answer is the city name or the product code, merge them.
6. Label where required, disclose where it helps. Merchant Center labelling is mandatory for AI product data. On the website, Google’s helpful content guidance asks whether the use of automation, including AI generation, is “self-evident to visitors through disclosures or in other ways”, and recommends accurate bylines where readers expect them.
7. Audit what is already live. Many sites already hold AI pages published in bulk over the last few years. Review them against sections 4.6.5 and 4.6.6 and improve, merge or remove the weakest.

What to do this quarter

For most companies, a short list is enough:

• List every page and product description created mainly with AI, in both languages (a technical SEO audit crawl gives you the full page list to start from).
• Check those pages’ titles, meta descriptions, structured data and alt text for accuracy.
• Identify groups of near-identical pages (cities, services, products) and decide which to enrich and which to consolidate.
• If you run Google Shopping, confirm your feed uses structured_title and structured_description for AI text and that image metadata survives your pipeline.
• Agree a written AI review rule for whoever publishes on your site, whether in-house or an agency.

This is the kind of review our SEO and search visibility team runs: a technical and content audit of what is live, then service and product pages rebuilt around what buyers in each market actually search for. When the problem is the structure of the site itself, it becomes website work. Either way, the October update gives you a concrete standard to measure your pages against.