What a brand-new company looks like to ChatGPT: Citeshine's day-0 baseline

Oct 13, 2026

Citeshine helps brands find and fix what AI gets wrong about their products. We’re especially strong with SaaS and AI tools selling overseas. We test global AI engines (ChatGPT, Perplexity, Gemini) and Chinese ones (DeepSeek, Doubao, Kimi). Start with a free sample check, upgrade to the full $99 audit when you want the fix list.

We do GEO (generative engine optimization): we help brands find and fix what AI gets wrong about them. We’re especially strong with SaaS and AI tools selling overseas. Our first case study is ourselves. This is what ChatGPT knew about Citeshine on October 13, 2026, and what we are doing about it. This is day 0 of building in public. We will publish every re-test and compare it with this baseline.

TL;DR

• In 11 generic buyer questions, ChatGPT mentioned Citeshine 0 times (0/11).
• In 2 questions that named us directly, ChatGPT described citeshine.com as a Shopify phone-case store registered in 2021, citing Merchant Genius. That was the domain’s previous owner. It has nothing to do with us.

So ChatGPT did not just know little about us. It knew nothing about us, and the little it “knew” was about someone else. A wrong answer is worse than a missing one, so our fix plan starts with correcting it.

Method

Limits, up front: n=1 means high variance. Logged-out web answers can differ from API answers and from logged-in or personalized ones. ChatGPT’s web UI shows sources as chips, so we recorded source names but no clickable citation URLs. This shows how ChatGPT answered on that day in that mode. It does not show what every user sees. ChatGPT web is also the only engine we have used for a published report so far; the other engines we plan to test are not covered here.

Results by prompt

Chinese prompts are translated here. The exact wording is in the GitHub report.

What ChatGPT said when asked by name

From the English answer (en06):

“The domain citeshine.com was reported as a Shopify-hosted e-commerce site registered on December 7, 2021… it may have sold phone cases/accessories…” (source chip: Merchant Genius)

The Chinese answer (zh07) said the same thing: a Shopify store record, domain registered December 2021, phone cases and accessories, source Merchant Genius. Both answers then gave shopping-safety advice, such as using a payment method with buyer protection.

For a GEO agency, that is worse than not being mentioned. A prospect who asks an AI about us gets a profile of a dubious web shop.

It also points to a trap many new brands fall into: domains have history. A second-hand domain, a renamed product, a name shared with another company: AI can attach the old information to you.

Who ChatGPT mentioned instead

Across the same 13 answers, GEO tools and agencies that appeared (answers mentioning them / 13):

First Page Sage, iPullRank, Siege Media, Animalz and Omniscient Digital also appeared (2/13 each). These are the tool’s raw substring matches on the full answer text, source labels included. Two consequences: one of the four 智推时代 / GenOptima hits is only a source label in an English answer, and two English answers that said “Directive” without “Consulting” were not counted. These are counts, not our opinion of these companies.

One pattern: when ChatGPT recommended providers, the source chips mixed providers’ own sites with roundup articles, media coverage and third-party lists (e.g. iThome, Jiemian News, Clutch). In the two Chinese answers that listed domestic GEO teams (zh01, zh05), every chip was iThome or Jiemian News, not a provider’s site. That fits the GEO paper’s description of generative engines that synthesize answers from several retrieved sources (Aggarwal et al., KDD 2024).

The fix plan

Already done (live on 2026-10-13, you can check):

1. Site is live and open to AI crawlers. Our robots.txt explicitly allows OAI-SearchBot, GPTBot, PerplexityBot, ClaudeBot and others. OpenAI says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers (OpenAI crawlers). Perplexity recommends allowing PerplexityBot (Perplexity crawlers).
2. We say plainly that we are not that store. Both llms.txt and the Organization structured data on the homepage state that Citeshine is not a Shopify store, does not sell physical products, and that citeshine.com previously belonged to an unrelated owner.
3. The tool, prompts and report are public. citeshine-audit (MIT), the prompt config and this run’s report are on GitHub, so anyone can re-run the same questions and compare.

Next:

4. Consistent entity info. The same one-line description and domain on the site, GitHub, X, Zhihu, Tencent Cloud Developer Community, Product Hunt and Reddit. Not done yet: some profile bios still use older wording. sameAs only lists accounts that actually exist.
5. Content an AI can quote. GEO vs SEO, a practical checklist, a tools comparison, each with method and sources. The GEO paper found that adding citations, quotations and statistics improved visibility in generative answers, while keyword stuffing did not help (Aggarwal et al.).
6. Third-party sources, earned honestly. Open-source usage, technical community posts, public data. No paid placements, no fake reviews, no negative content about competitors.
7. Re-test with a fixed prompt set. Same 13 prompts, same logged-out web mode, at least 3 runs per prompt, plus Perplexity and DeepSeek via API. We will also check Chinese AIs such as Doubao and Kimi and say how each result was collected. We will publish the report for every re-test, including ones where nothing changed or things got worse.

What we will track:

• When asked by name, does ChatGPT correctly say Citeshine is a GEO service? (Today: 0/2 correct.)
• Mentions across the 11 generic questions. (Today: 0/11.)
• Answers citing citeshine.com or our GitHub. (Today: 0/13.)

We are not promising numbers or dates. AI answers vary from run to run, and nobody can guarantee that ChatGPT will recommend you. What we can do is publish verifiable information and keep measuring it the same way.

Why this is more than asking ChatGPT once

Asking ChatGPT one question takes ten seconds. Finding out why it is wrong, fixing it, and proving the fix held is the actual work. This post is only step one.

1. Trace the error to its source. “ChatGPT got it wrong” is not actionable. “ChatGPT says citeshine.com is a Shopify phone-case store registered in December 2021, and cites Merchant Genius” is. Each wrong answer has to be traced through its cited sources to the page the claim came from, and the fix depends on what that page is: your own site, a data aggregator, a directory, an old article.
2. Repeat runs, not single answers. The same question can get a different answer a minute later. One answer proves nothing, positive or negative. A useful measurement fixes the prompt set, runs each prompt several times, and reports rates with n, engine, mode and date. This baseline is one run per prompt; for re-tests we plan at least 3 runs per prompt.
3. Overseas and Chinese AIs are different problems. ChatGPT, Perplexity and Gemini on one side, Doubao, Kimi and DeepSeek on the other, use different crawlers, different indexes and different sources, and cite them differently. A fix that works for one may do nothing for another. So far we have only tested ChatGPT web; the others are next, and we will say how each result was collected.
4. Fix, then re-test to prove it. A change counts only when a re-test with the same prompts shows the answer changed. Every re-test adds to a history for the same brand and the same questions, which is what lets you tell a real improvement from noise. That history only builds up over time.

Why teams rarely do this themselves: it is not hard once, it is hard to keep doing. Fixed prompts, repeated runs on six engines in two languages, tracing sources, contacting third-party sites, and re-running every few weeks is a recurring job that usually falls behind product work. That is the part Citeshine is built to do, and we are starting with our own brand, in public.

Want to check your own brand?

• Do it yourself: run citeshine-audit with 10–20 questions your buyers actually ask, at least 3 runs each, and always record engine, mode, date and n.
• Or have us do it: we’re especially strong with SaaS and AI tools selling overseas. We check global and Chinese AI engines and start with what they get wrong about you. The audit is free. If we work together afterwards, you pay a monthly base fee plus a bonus only when a re-test shows the metric reached the target we agreed on in writing. citeshine.com/audit

Find and fix what AI gets wrong about you.