Multilingual SEO Best Practices: A Practical Guide

Alfredo BernalOrganic Growth LeadatOmnia
Alfredo BernalOrganic Growth LeadatOmnia
TL;DR
Your founder types a question about your category into ChatGPT. The answer recommends a competitor and cites three pages. None is yours, and you don't know what to change.
A site-wide checklist won't fix that. ChatGPT cites pages for a specific prompt, so the fix starts with one prompt and one page. The six steps below take you from choosing the prompt to checking, weeks later, whether the citation arrived. The same steps show how to get cited by AI engines other than ChatGPT, once you run step 2 for each one.
A citation here means ChatGPT pointing to your page as a source. For the full definition, read our entry on AI citation optimization. For the brand-wide view, see our brand visibility playbook. This guide covers a single page.
You need a prompt to work on, a page you can edit, and a way to see what ChatGPT cites.
Check one technical condition as well. OpenAI says a site should allow OAI-SearchBot in its robots.txt file to appear in ChatGPT search answers. It adds that robots.txt changes can take about 24 hours to affect search results (OpenAI crawler documentation, no date shown, checked 4 November 2026).
For one SaaS prompt, ChatGPT cited 76 distinct third-party pages and no social ones between 5 October and 4 November 2026.
Omnia monitors the US prompt "How do I improve visibility in ChatGPT for a US SaaS brand?" daily. Over those 30 days, ChatGPT's answers cited 76 distinct third-party URLs and no social URLs. The most cited were OpenAI's own pages about its web crawlers, with four citations each for two of them. Beyond those, the top ten included an Adobe blog post, two research reports, an arXiv survey and an Axios article.
| Source type | Distinct URLs ChatGPT cited | Example |
|---|---|---|
| Third-party | 76 | OpenAI's crawler documentation |
| Social | 0 | None cited |
That is one prompt, one engine and 30 days of daily runs, counted as citations rather than answers. Read it as a snapshot of what ChatGPT reached for on this prompt, not as a rule about what it cites. The takeaway: the pages ChatGPT cites for your prompt are your reference set, and every step after the first works from that set.
Pick one prompt written the way a buyer talks to ChatGPT: a full sentence with their context, not a keyword. A keyword is what you type into Google. A prompt is what you say to an AI engine.
Say you sell invoicing software. Two keywords turn into prompts like this:
Seed your prompts from the questions buyers ask on sales calls and in support tickets, then phrase them as sentences. Choose one prompt per page, so you know which answer you are trying to win.
Run the prompt in ChatGPT and write down every URL it cites, and whether each is owned, third-party or social.
Doing this by hand works for one prompt. Omnia runs your prompts daily in real browsers, stores each full answer and lists the cited URLs by type, so you don't screenshot anything. For the longer method, see our guide to tracking AI citations and our comparison of citation analysis tools.
Compare where each cited page puts its answer, what evidence it gives, how it is structured and how recently it was updated. The table lists the four checks.
| Check | What to look for | What a gap looks like |
|---|---|---|
| Answer position | The first sentence answers the prompt | Your page reaches the answer in paragraph four |
| Evidence | Figures with a named source and a date | Your page makes claims with no source |
| Structure | Headings that match sub-questions, a table or list where it fits, a visible FAQ | One long block of prose under a vague heading |
| Freshness | A visible updated date and current figures | Your page was last updated before the topic changed |
The comparison won't tell you why ChatGPT picked a page. It tells you what the cited pages have that yours lacks, and that is the part you can change. Write down the two or three biggest gaps. They are your edit list.
Rewrite the page so its first sentence answers the prompt and every section makes sense when lifted out on its own. Meltwater, in a case study published on 4 August 2026, lists six patterns in its most-cited pages. They include self-contained sections, direct answers at the top of each section, and claims backed by specific figures and named sources.
To improve AI citations for your prompt, make these five edits:
Schema won't rescue a weak page. Meltwater says it helps machines parse a page's structure but does not create citable content by itself.
The Omnia Agent can draft the rewrite from your brand's own citation data. Omnia does most of the work; you make the calls and approve what ships.
If most pages ChatGPT cites belong to other sites, ask those sites to include you. Your step 2 list shows which ones: roundups, review pages, research reports and trade articles.
In the prompt above, the third-party pages beyond OpenAI's own documentation included original research and a trade article. If you hold data other sites can cite, publish it and send it to the authors of the pages already on your list.
Three actions are worth the time. Pitch inclusion on the roundups ChatGPT cites for your prompt. Offer a figure or a quote the author can use. Keep your listing accurate on any review page in your list.
The Omnia Agent can draft the pitches, and nothing is sent until you approve it.
Run the same prompt for several weeks and compare which URLs ChatGPT cites before and after the change. Keep the baseline from step 2 and judge the trend, not a single day.
Answers vary between runs, so a citation that appears once and disappears tells you little. A URL that keeps appearing across weeks tells you the edit worked. Omnia monitors your prompts daily, so the before-and-after comparison builds itself, and it reports which URLs were cited, not why the engine chose them.
If nothing changes after several weeks, go back to step 3. The gap you fixed may not be the one that mattered, so pick the next one on your list and change only that.
The most common mistake is optimizing a page without knowing which prompt it should answer. Go back to step 1 and write the prompt first.
Judging from one run is the next. Answers vary, so one answer proves nothing. Run the prompt on several days before you decide what to change.
Changing many pages at once makes the result impossible to read. Change one page, keep the baseline, and you can say which edit moved the citations.
Treating schema as the fix wastes the week. Structure helps machines read the page, but the answer and the evidence are what a citation needs.
Pick one prompt this week and write down what ChatGPT cites for it. That list is your baseline.
Omnia shows which URLs ChatGPT, Perplexity, Google AI Overviews and Google AI Mode cite for each of your prompts, daily. The Omnia Agent drafts the page update for you to approve. Start with the free ChatGPT visibility checker, then start your 14-day free trial, no card needed.
Omnia offers a 14-day free trial on the Growth plan.
No credit card required. See exactly where your brand shows up (or doesn't) across AI engines, then let the platform's recommendations guide your next move.
There is no fixed time. Meltwater reported citation growth within about five weeks of its site relaunch going live, and said timing varies with how often each engine re-crawls and re-indexes content. Track weekly, not daily.
Yes, but run step 2 for each engine, because the cited pages differ. Omnia monitors ChatGPT, Perplexity, Google AI Overviews and Google AI Mode by default.
For the prompt measured above, ChatGPT cited no social URLs between 5 October and 4 November 2026, while the other engines cited Reddit 11 times for the same prompt. That is one prompt, so check yours in step 2.
Read what llms.txt is, and whether you need one, in our explainer on llms.txt.
No. Brand visibility is the brand-wide work of being mentioned and cited across many prompts. This guide covers one page and one prompt.
Written By

Alfredo BernalOrganic Growth LeadatOmniaNo credit card required · Free for 14 days · See your AI visibility within 3 minutes