Sitegravity
← All research

We tested ourselves: Sitegravity's mention rate in AI answers is 0%

Published: November 12, 2026

We ran our own test on ourselves in Malaysia and Macau. Sitegravity was named in none of 432 AI answers. Here is how we tested, what the businesses that were recommended have in common, and what we will do next.

Why we tested ourselves

We tell businesses how often AI assistants name them. It would be strange to do that without checking our own number. So we ran the same kind of test on Sitegravity, in the market we sell to: businesses that want to be recommended by AI in Malaysia and Macau.

How we tested

We wrote 28 realistic questions of the kind a business owner or marketing manager asks: which companies can check and improve how AI describes my business, what GEO is, how it differs from SEO, how much it costs. Malaysia got 16 questions (4 each in English, Simplified Chinese, Traditional Chinese and Malay) and Macau got 12 (4 each in English, Simplified Chinese and Traditional Chinese). Each question was asked three times on six AI engines: ChatGPT, Gemini, Perplexity, DeepSeek, Doubao and Yuanbao (the last two with Chinese questions only). That gives 240 answers for Malaysia and 192 for Macau, 432 in all, collected on 9 November 2026. No question contained our name or any company name.

Result: 0 of 432

Sitegravity was named in none of the 432 answers: not on any engine, in any language, or for any type of question. We also searched the full text of every answer for our name in its spellings, and found none. No answer cited our website as a source.

We did not expect to be visible. The site went public in November, and AI engines that search the web take weeks to months to reflect new pages. The number matters because it is our starting point: 0%.

What the businesses that were recommended have in common

The questions asked for providers, so the answers were full of names. We do not name any of them here. What we looked at is why those businesses were named.

  • Their own websites were the sources. Of the 22 most-cited websites in each market, at least 17 in Malaysia and at least 12 in Macau were the providers' own sites. The pages cited were mostly service pages and explanatory articles ("what is GEO", "how much does it cost") written by the provider, and some were "best agencies" style lists on their own sites.
  • Directories and neutral explainers were cited too. One business-directory site with "top 10" pages and company profiles was cited in 44 of 240 Malaysian answers (18%). Industry news sites, search-engine documentation and research papers also appeared among the most-cited pages.
  • DeepSeek and Yuanbao name the most businesses. In the Malaysian answers, DeepSeek named at least one business in 96% of answers and Yuanbao in 92%, against 69% for ChatGPT, 69% for Gemini and 46% for Perplexity. In Macau: DeepSeek 81%, Yuanbao 83%, Gemini 64%, ChatGPT 61%, Perplexity 56%.
  • The question type matters, differently in each market. Questions that asked for a recommendation almost always produced names (96% to 97% of answers). In Macau, price questions named a business in 73% of answers, against 17% for "what is GEO" questions and 44% for comparisons. In Malaysia price questions did so in only 27%, so we do not draw a general rule about price questions.

What we are doing next

  1. Writing the answers we want AI to find. We have published three guides on our own site: what GEO is, how GEO differs from SEO, and how GEO services are priced.
  2. Getting listed correctly on the business directories that AI answers cite.
  3. Telling search engines about our pages. We now submit our sitemap through IndexNow, which Bing and other search engines use.

None of this will change the number overnight.

Our commitment

For the next three months, we will publish the result of this same test, with the same 28 questions, once a month. If the number stays at 0%, we will say so. Next result: December 2026.

Monthly updates

The December result will be published in December 2026.

FAQ

Why publish a bad number?

Because it is true, and because we ask other businesses to measure theirs. A measurement is only useful if you report it when it is unflattering.

Did you use the same method as for clients?

Yes: realistic questions, several engines, several languages, three repeats, and every answer analysed. The only difference is that the questions were about our own kind of business.

Why not name the businesses that were recommended?

This is a test of ourselves, not a ranking of other businesses. What matters for any business is what those answers have in common.

Method and limits

One test on one date; AI answers vary from one attempt to the next, and models change. We call the engines through their APIs, not the consumer apps; Doubao ran without web search. "Named a business" counts businesses the answer recommended or that we track, found by our automatic entity extraction, and leaves out AI products and general platforms; it has not been checked by hand. "Providers' own sites" was judged from the web addresses of the cited pages, and both figures are lower bounds because some sites in each market could not be classified. Our sampling method and its limits are on the Method page.

Data collected November 9, 2026. Findings are anonymised and describe one test on one date. → Method

Want to see this for your brand?

Get a free AI visibility check