Why Adding English Pages Could Boost Your AI Citations

What changed in English pages and AI visibility?

English pages appear to have become a stronger retrieval surface for some AI answer systems, even when a user asks in another language. Search Engine Land published a multilingual-site analysis on August 5, 2026 that examined data from ChatGPT, Copilot, Google AI Mode, and AI Overviews. Its central finding was not that every engine behaves alike. It was that ChatGPT showed a pronounced preference for English pages during retrieval, while the effect was far smaller in Copilot and Google AI surfaces.

The analysis separates traditional search demand from AI citation behavior. It measured English over-indexing as English's share of AI citations divided by English's share of traditional search impressions. A score of 1.0 is neutral. In the reported dataset, Copilot was near neutral at 1.07, Google AI was below neutral at 0.79, and ChatGPT was roughly 2.6. That comparison matters because it points to a retrieval pattern rather than a universal rule about multilingual search.

The reported log evidence also distinguishes OpenAI agents. GPTBot is associated with content that may be used for model training. OAI-SearchBot is used to surface websites in ChatGPT search features. ChatGPT-User may visit pages for a user-initiated request, but OpenAI says it is not used to decide whether content appears in Search. This is why a robots review should focus on the crawler that affects the visibility surface you want, rather than treating every OpenAI user agent as interchangeable.

The practical change is simple: an English version of a core page can now be the page an AI system encounters and cites. That can happen even when the local-language version remains the page that earns most human traffic. A multilingual content strategy must therefore consider whether the English page would be safe and accurate if it became the answer source for a customer anywhere in the world.

1. Before August 5, 2026: English translation work could be framed mainly as a conventional international-search decision.
2. Observed on August 5, 2026: the cited analysis reported an English retrieval bias that was strongest in ChatGPT.
3. What to do now: audit whether important English pages are complete, current, crawlable, and aligned with their source-language equivalents.
4. What not to do: publish a thin English folder simply to chase an AI citation.

Who does this affect most?

This affects multilingual companies whose customers use AI assistants to research products, services, pricing, regulations, or local options. B2B SaaS teams with French, Dutch, German, Spanish, or other local-language sites are exposed when a buyer asks an assistant for a category recommendation and the assistant retrieves an English counterpart. Multi-location service businesses face the same issue when an AI answer combines local intent with broader web retrieval.

The clearest exposure is not a site that lacks English pages. It is a site with a neglected English footprint. The reported analysis describes a bank whose outdated English rates PDF remained online after Dutch and French pages changed. If an AI system retrieves that document, the user receives an outdated answer regardless of the bank's current local-language information. Citation visibility without content governance becomes an accuracy liability.

Teams with large documentation libraries should also pay attention. Product pages, pricing explanations, help-center articles, comparison pages, policy documents, and downloadable assets can all create contradictory English sources. The risk rises when localization is owned by one team, content updates by another, and technical publishing by a third. A page does not need high organic traffic to become a source an AI system retrieves.

The effect may be particularly important for companies that already see citations but cannot explain which URLs are being used. Citation Share tells you the percentage of relevant AI answers in a category that cite you. It does not replace URL-level inspection. A brand can gain Citation Share while losing message control if an obsolete English page becomes its most visible evidence.

1. Multilingual B2B companies with category and comparison content.
2. Local businesses serving residents, visitors, or cross-border customers.
3. Organizations with legacy PDFs, old product pages, or duplicated help content.
4. Teams measuring AI answer presence without checking the exact cited URL.

Does an English folder actually increase AI citations?

An English folder was associated with higher AI visibility in the reported analysis, but it is not proof that adding one will cause the same result for every website. The data compared sites with an English folder against sites without one and normalized AI visibility against traditional impressions. It is an observed relationship across the study panel, not a guarantee for an individual domain.

The strongest reported difference appeared in ChatGPT-related bot data. Sites with an English folder saw a 122% uplift. In Copilot, sites with English pages received 892 citations per 10,000 impressions compared with 585 for sites without them, an aggregate difference of 52%. In Google's Gen AI panel, the reported aggregate lift was 28%; after outliers were removed, the median per-site lift was 9%.

Those figures give an editorial team a useful before-and-after decision frame. Without a maintained English retrieval surface, a multilingual site sat at the baseline numbers above. With one, reported AI visibility rose across the surfaces studied. That does not call for duplicating every URL. It calls for building a controlled English source around the pages that answer high-stakes questions and can be kept current.

The study cannot establish whether the English folder alone caused every lift. Companies that invest in English localization may also have stronger technical SEO, larger content operations, more complete product information, or better authority. Use the figures as a reason to test and measure, not as a promise to stakeholders.

1. No English folder: 585 Copilot citations per 10,000 impressions in the reported comparison.
2. English folder: 892 Copilot citations per 10,000 impressions in the reported comparison.
3. Observed aggregate difference: 52% more Copilot citations per 10,000 impressions.
4. Recommended response: publish and measure a maintained English cornerstone-page set before expanding further.

How should a multilingual site respond?

A multilingual site should begin with a narrow English content audit, then publish only the pages it can govern as durable sources. Start with the pages that shape a buying decision or create customer risk: category pages, product overviews, service descriptions, pricing explanations, comparison pages, policy pages, and critical documentation. Ranking potential is not the deciding factor. What matters is whether the page would be accurate enough for an AI assistant to cite tomorrow.

For each selected URL, compare the English page with its primary-language counterpart. Confirm that claims, pricing context, product availability, regulatory wording, dates, and calls to action remain aligned. Check downloadable PDFs and help-center entries as carefully as HTML pages. A stale asset can remain retrievable long after a navigation update hides it from ordinary visitors.

Then establish the technical relationships. Google recommends explicitly indicating localized versions so it can point users to the appropriate language or region version. Its documentation says localized variants can be marked using HTML, HTTP headers, or a sitemap, and that reciprocal alternate links are important. Hreflang does not itself make a page eligible for an AI citation, but it helps preserve language and locale relationships in conventional search systems.

Check crawl controls before you declare success. OpenAI states that OAI-SearchBot is used to surface websites in ChatGPT search and recommends allowing it in robots.txt for sites that want to appear there. Verify the relevant robots rules, server response, canonical URL, indexability, and page rendering. Do not conflate GPTBot permission with ChatGPT search visibility, because OpenAI documents them as distinct controls.

Finally, measure the test as a citation program. Build a prompt set in the languages your customers use. Record the answer language, cited domain, cited URL, engine, answer presence, and date. Track Citation Share across relevant prompts. Review the English pages that appear most often, then refresh them alongside the primary language. That cycle turns English localization into Citation Engineering rather than a one-off translation backlog.

1. Audit English pages that affect customer decisions or factual accuracy.
2. Remove, update, redirect, or clearly replace obsolete English files.
3. Publish fully maintained English versions of priority pages.
4. Implement and validate language relationships, crawl access, canonicalization, and rendering.
5. Measure citations by engine, language, URL, prompt, and date.

What should you avoid when adding English pages?

You should avoid publishing English pages that are thinner, older, or less precise than the source-language pages they replace. AI retrieval does not reward a site for having more URLs if the retrieved page gives an incomplete answer. A weak translation can damage trust because it may become the language version a prospective customer sees first.

You should also avoid declaring English the canonical business language when it is not. The goal is not to replace local expertise with generic English copy. It is to make the authoritative version of key information available in English while preserving the correct regional details. That requires decisions about pricing scope, supported markets, legal wording, and local service availability.

Do not roll out an English folder without an owner and update path. Every translated cornerstone page needs a source page, a responsible team, a review cadence, and a rule for what happens when the source changes. Without that operating model, the folder becomes a growing collection of unsupported claims.

Avoid measuring success with raw citation count alone. Citation Count per day can show volume, but it cannot tell you whether the citations came from the right pages or covered the questions that matter. Pair it with Answer Presence and Citation Share. Review the wording around the citation so you can see whether the assistant understood the page accurately.

1. Do not ship machine-translated copy without subject-matter review.
2. Do not leave English PDFs outside the regular content-refresh process.
3. Do not block the relevant search crawler by accident.
4. Do not promise a citation increase from a correlation study.
5. Do not count citations without reviewing the cited URL and answer context.

Sources

Source: Search Engine Land

Multilingual-site analysis reporting English retrieval bias and English-folder visibility comparisons across ChatGPT, Copilot, and Google AI surfaces. Search Engine Land, 2026-08-05

Source: OpenAI Developers

OAI-SearchBot is used to surface websites in ChatGPT search features; GPTBot and ChatGPT-User have separate documented roles. OpenAI Developers, 2026-09-04

Source: Google Search Central

Google recommends explicitly indicating language and regional variants, including reciprocal alternate relationships, to help serve the appropriate localized page. Google Search Central, 2026-09-04