How to show up in ChatGPT: what three weeks of citation tests showed us
· Nolyn research team
How do you show up in ChatGPT answers?
Publish a page that answers one specific question with dated data (a list with its date window and sample size), worded the way a buyer asks, and keep it current. In our logged-out ChatGPT checks from October 17 to November 10, 2026, dated lists were the pages it cited most; one was cited about 10 hours after going live, before Google or Bing had crawled it, while six new tool and calculator pages checked on launch day were cited 0 times in 40 runs.
This guide is for brands, agencies and e-commerce teams who want their pages to appear as sources in ChatGPT's answers (often called ChatGPT SEO or LLM SEO). ChatGPT is our largest source of visitors, so since October 17, 2026 we have asked ChatGPT the questions our pages answer and recorded what it cited. Everything below comes from those checks and from our own analytics, with the date and sample size next to each claim. The samples are small and ChatGPT's answers change from run to run, so read the patterns as evidence from one site, not as rules or a promise of rankings. We never publish anyone's account or visit data; traffic figures are aggregates.
How we tested
We opened chatgpt.com logged out in a clean browser, asked a question through ChatGPT's search mode, waited about 50 seconds for the answer, and recorded the cited hosts, whether Nolyn appeared, and whether ChatGPT searched the web at all. Logged out means no personal memory or chat history shapes the answer. A question counts as cited only when a nolyn.ai source chip or link is visible; chips that read "+2" hide sources we could not open logged out, so a few citations may be missed. Our test browser exits in Japan, which sometimes pulls answers toward Japanese sources.
Between October 17 and November 10, 2026 we ran more than 150 such checks in English, Portuguese, Japanese and Spanish. Most questions were run once; where a question was run two or three times, the runs are counted separately below. ChatGPT gives different answers to the same question on different runs, so a single run is weak evidence; the patterns below held across several dates.
| Date | Questions | Nolyn cited | What was cited |
|---|---|---|---|
| Oct 17 | 6 broad tool and trend questions (pt, en) | 0 of 6 | None; vendor pages and a competitor's weekly sales list |
| Oct 18 | 15 broad questions (pt, en) | 2 of 15 | A pt engagement-rate calculator and a pt how-to; 0 on "best tools" questions |
| Oct 22 | 10 seller questions (pt, en) | 4 of 10 | Shopee fee calculator and guide, a Fakespot alternative page, an Amazon Brasil best-seller list one day after launch |
| Oct 24 | 10 seller how-to and tool questions (en, pt, ja) | 0 of 10 | None; roundups and official help pages |
| Oct 25 | 10 US seller and brand questions (en) | 3 of 10 | Influencer finding, free social listening tools, private label products; all three pages one day old and not yet indexed |
| Oct 29 | 10 en + 10 ja seller questions | 2 of 10 each | Dated product lists; 0 on calculator, fee and method questions |
| Oct 31 | 9 video-analysis questions (en, es, pt, id) | 4 of 9 | The TikTok video analyzer page |
| Nov 1–3 | 6 new tool and comparison pages, 40 runs on launch day | 0 of 40 | None; free no-signup tools and vendor listicles |
| Nov 10 | 9 questions re-run for this guide (pt, en) | 2 of 9 | The November beauty list and the Black Friday page, on a seller-phrased question |
Nolyn's logged-out ChatGPT checks, 2026-10-17 to 11-10. Cited = a nolyn.ai source visible in the answer.
What ChatGPT cites: dated, specific answers
The pages ChatGPT picked up again and again were lists with a date window and a count: "products with the most TikTok Shop video views in Brazil, October 13–20", "trending pet products", "best-selling products on Amazon Brasil". When ChatGPT cites such a page it tends to copy the structure almost verbatim: the ranked items, the figures, the date window, and even our caveats. On October 20 it reproduced a Top 10 from one of our Brazilian pages with the line "views are not sales"; on November 10 it answered "produtos de beleza em alta no TikTok Shop Brasil" with our November list (October 3 to November 1 window, view counts per product) and named the source as "levantamento Nolyn".
What these pages share is that they answer a question nobody else has answered with fresh numbers. ChatGPT seems to search for the most specific, most recent answer it can quote, and a dated list is easy to quote. Our Brazilian beauty list was cited in 3 of 3 runs on October 21; our US pet products list was cited within 12 hours of launch on October 29.
What it passes over
Four kinds of question rarely led to Nolyn, even when we had a page that answered them:
- "Best X tools" and "X alternatives". ChatGPT cites vendor homepages and listicles. "Best Kalodata alternatives" cited us 0 times on October 17, October 18 and November 10; our two ad-spy alternatives pages were 0 of 4 on launch day.
- Market statistics. "TikTok Shop Brazil statistics" was answered from TikTok's newsroom, consultancies and the press (users, GMV, store counts) in 0 of 5 runs from October 21 to November 10, although our statistics page answers a different, video-level layer.
- Calculators. Two calculator pages were cited in October (an engagement-rate calculator on October 18, a Shopee fee calculator on October 22), but calculator and fee questions cited us 0 times on October 29, in English and Japanese, and our TikTok money calculator was 0 of 12 on launch day. For calculation questions ChatGPT often builds its own calculator in the answer instead of linking to one; on November 2 it did this in every calculator phrasing we tried.
- How-to questions with an official answer. Fees, policies and setup steps go to the platform's own help pages (TikTok Seller, Sell on Amazon, Sebrae in Brazil). A how-to page only got cited when it carried something the official page does not, such as our own counts.
How the wording of the question changes the answer
The same topic gets different sources depending on who the question sounds like. A generic Black Friday question reads like a shopper's, so ChatGPT answers with the sale calendar and coupons from news sites; the seller's version of the same question cited our Black Friday page both times we asked. Some phrasings trigger no web search at all: when asked to compare a TikTok video with the creator's average views, ChatGPT offered to do the maths itself in all 4 runs and cited nothing.
So before you write a page, check which wording your buyers actually use and whether that wording makes ChatGPT search. Words that point to data ("2026", "this month", "benchmark", "what to sell") made it search more often in our checks; requests to judge or calculate something made it answer from its own knowledge.
| Question as asked | Nolyn cited | What ChatGPT used instead |
|---|---|---|
| Black Friday TikTok Shop Brasil 2026 | 0 of 5 | Sale dates, coupons and discounts from UOL Economia, InfoMoney, E-Commerce Brasil |
| o que vender na Black Friday do TikTok Shop Brasil 2026 (what to sell) | 2 of 2 | Our October 13–20 product video list, next to a competitor's 7-day category sales |
| estatísticas TikTok Shop Brasil | 0 of 5 | TikTok newsroom, Momentum Works, BTG Pactual |
| compare a TikTok video to the creator's average views, is it performing well | 0 of 4 | No web search; ChatGPT offered to compute it |
Same topic, different phrasing. Logged-out ChatGPT, runs on 2026-10-21 and 11-10 (the comparison question on 11-02 and 11-10).
How fast a new page can be cited, and how fast it goes stale
A page does not need to be in Google or Bing first. On October 21 our Brazilian beauty list was ChatGPT's main source about 10 hours after it went live, while both Google and Bing still listed it as discovered but not crawled. On October 25, three US pages were cited one day after launch, also before indexing. OpenAI runs its own search crawler (OAI-SearchBot), and pages it can fetch can be cited without a search-engine index, at least in our cases.
Minutes after launch is too soon, though. Six new pages checked between 5 minutes and about an hour after deploy (November 1–3) were cited 0 times in 40 runs. A retrieval test on October 25 suggested ChatGPT reads a copy of the page that is at least half a day old.
Dated pages also expire. On October 31 and November 1 our ChatGPT visitors fell by roughly 45–70% across Brazil, the US and Japan; ChatGPT began framing "this month" answers as November, and for at least one question switched to sources dated November. We refreshed the dated pages with November data; ChatGPT visitors went from 44–50 a day (October 31, November 1) to 77 on November 2, and the refreshed pages were cited again on November 3. The US pet products list, published October 28 and not refreshed, was cited on October 29, but on November 10 no Nolyn source was visible and the answer was built on sources labelled November 2026. If a page's value is its date, update it before the month turns.
What ChatGPT traffic looks like: 30 days of our analytics
From October 11 to November 10, 2026, 1,670 of our 4,544 visitors (37%) arrived from chatgpt.com, more than Google (656). Visits recorded with a chatgpt.com source also included 383 completed sign-ups, against 54 for Google. Weekly ChatGPT visitors grew from 421 (week of October 19) to 523 and then 606 (week of November 2). By language, 786 landed on English pages, 776 on Portuguese and 107 on Japanese.
Dated case studies brought the most ChatGPT visitors, but tool pages that read live data brought the most sign-ups. Our logged-out checks predicted tool traffic poorly: the Instagram Reels analyzer was cited 0 of 12 times at launch yet had 16 ChatGPT visitors in its first week, and a question about free Instagram audit tools cited other vendors on November 10 while our Instagram audit page is one of our most visited from ChatGPT. We cannot see the questions people typed, so measure ChatGPT referrals per landing page alongside any citation checks.
| Landing page type | Pages landed on | Visitors | Sign-ups |
|---|---|---|---|
| Case study with dated data | 46 | 811 | 104 |
| Data tool (reads live social or marketplace data) | 47 | 494 | 222 |
| Guide | 21 | 131 | 11 |
| Calculator | 12 | 120 | 8 |
| Home page | 4 | 74 | 30 |
| Comparison or alternatives | 13 | 40 | 4 |
Visitors (distinct browsers) and completed sign-ups in visits whose source was chatgpt.com, 2026-10-11 to 11-10, by landing page type. A browser can land on more than one type.
What we could not prove
Several common ChatGPT SEO claims we could neither confirm nor rule out with our data:
- Whether llms.txt helps. We publish one, but we have no page without it to compare against.
- Whether structured data or answer-first layout changes citations. We moved the short answer above the form on our tool pages on November 7, 2026; it is too early to measure the effect.
- Whether Bing's index matters at all. Our pages were cited before Bing crawled them, but that does not show ChatGPT never uses Bing.
- What logged-in users see. Memory, location and chat history change answers; our checks are logged out and exit in Japan.
- Which questions real visitors asked. ChatGPT passes the source (utm_source=chatgpt.com) but not the question.
- Anything about Perplexity, Gemini or Google AI Overviews. Perplexity sent us 8 visitors in the same 30 days, too few to study.
How to check and improve your ChatGPT visibility in 7 steps
Write down 10 questions your buyers ask
Use their wording ("what to sell for Black Friday", "find influencers for my skincare brand"), not your category name. Mix list, comparison, how-to and "right now" questions.
Ask ChatGPT logged out, two or three times each
Use a private window and ChatGPT search. Record whether it searched the web, which sites it cited, and whether you appear. Answers vary between runs, so one run proves little.
Look at what won each question
Official help page, vendor listicle, news site, or a dated list? If ChatGPT did not search at all, rephrase the question toward data before writing a page for it.
Publish the specific answer with a date and a sample size
Put a two- or three-sentence answer at the top with one concrete figure and its date window. A list with counts is easier to quote than prose.
Make sure OpenAI's crawler can read the page
Allow OAI-SearchBot in robots.txt (OpenAI says changes take about 24 hours), return the content in the HTML, and avoid putting the answer behind a sign-in.
Re-check after a day, not after five minutes
In our tests pages were cited 10–24 hours after launch, never minutes after. Re-run the same questions a day later and again a week later.
Refresh dated pages before the month turns and track referrals
Update the data and the date before "this month" questions move on, and count visitors with utm_source=chatgpt.com per landing page; that is a better signal than any single citation check.
Questions
Partly. Crawlability and clear answers matter to both, but in our tests ChatGPT cited pages that had no Google ranking yet, including one cited about 10 hours after launch, and it favoured specific dated lists over broad, authoritative guides. Our site has a very low authority score, and ChatGPT still cited us in 2 to 4 of 10 questions on several dates.
In our checks, between about 10 and 24 hours for pages that got cited at all. Pages checked 5 to 60 minutes after launch were cited 0 times in 40 runs (November 1–3, 2026). Many pages are never cited for the question we targeted.
Not in our cases: a page was ChatGPT's main source while Bing and Google still listed it as not crawled (October 21, 2026). OpenAI runs its own search crawler, OAI-SearchBot. We cannot rule out that Bing helps in other cases.
Ask your buyers' questions in ChatGPT search while logged out, two or three runs each, and record the cited sources. Then compare with the visitors your analytics attribute to chatgpt.com per landing page, because checks and real traffic often disagree.
No tool can guarantee it. ChatGPT's sources change between runs and between days, and in our checks the same question cited us on one date and not on the next. A ChatGPT SEO tool can show you where you appear and which pages win instead; the page itself still has to be the most specific answer.
In our data the most common reason was age: when the month turned, ChatGPT moved to sources dated in the new month, and our visitors from ChatGPT fell by roughly 45–70% for two days until we refreshed the data. Run-to-run variation is the other reason, so check two or three runs before concluding.
Sources
- Nolyn: logged-out ChatGPT search checks on chatgpt.com, 2026-10-17 to 11-10 (more than 150 runs; question, date, cited hosts and whether a web search ran, logged per run) · checked November 10, 2026
- Nolyn: first-party visit and sign-up events for visits whose source was chatgpt.com, 2026-10-11 to 11-10, aggregated by landing page type (no individual data) · checked November 10, 2026
- Nolyn: dated TikTok Shop Brasil beauty list, the page ChatGPT cited before Google or Bing had crawled it (refreshed with November data) · checked November 10, 2026
- OpenAI: Overview of OpenAI Crawlers (OAI-SearchBot for ChatGPT search, ChatGPT-User, GPTBot; robots.txt changes take about 24 hours) · checked November 10, 2026
- Nolyn robots.txt (allows all user agents outside /api, /admin and sign-in pages) · checked November 10, 2026
Influencers and UGC (brands and agencies)
All guidesGuide
How to find influencers for your brandHow to find influencers for your brand: 6 places to look, a vetting checklist and a real search (40 results, 9 in range, 2 worth contacting), plus UGC creators.
Comparison
Best UGC platforms for brands in 2026Best UGC platforms for brands checked Oct 24, 2026: Insense, Trend, JoinBrands, Collabstr, Billo and more, with prices per video, usage rights and free options.
Guide
How to find micro influencersHow to find micro influencers in 7 steps, with a real search: 40 TikTok accounts, 9 in the 10K–100K range, 2 active US creators. Free and paid methods.
Comparison
Free social listening tools in 2026Free social listening tools checked on each vendor's site on Oct 24, 2026: 5 are free for good, the rest are trials. Limits, cheapest paid plans, AI features.
Comparison
GRIN alternatives in 2026GRIN alternatives checked on each vendor's site Oct 22, 2026: Upfluence, Aspire, Modash, Heepsy, Shopify Collabs and more, with prices and best fit.