Using Press Releases, note and Q&A Sites for ChatGPT Citations: Why Up to 42.8% of ChatGPT Answers Cite a Press Release in Japan, and What Gets a Release Cited
目次
Key Takeaways: In Japan, ChatGPT Optimization Runs on Three Channels: Press Releases, Long-Form First-Hand Accounts, and Q&A
Import a Western ChatGPT playbook into Japan and one thing always falls through the gap: PR TIMES, the country's dominant press release wire. Japan is the market we track most closely, and the mechanics here, a press wire the AI trusts, a long-form platform full of first-hand accounts, a giant Q&A site, have equivalents in most markets. Here's what I confirmed from Ahrefs' Japan data and two 2026 studies of press release citations.
- Up to 42.84% of ChatGPT answers cite a press release — Wallabee (Optyino.ai) analyzed 81,895 answers across nine AI models from April 21 to July 8, 2026. 42.84% of GPT-5 mini answers and 27.33% of GPT-5.2 answers cited at least one press release. Perplexity: 21.45%. Claude: 6.00%. Copilot: 2.90%
- 60.5% of cited press releases are on PR TIMES — Of 9,381 press release citation URLs in the same study, 60.5% were on the PR TIMES domain. In Ahrefs' Japan data, PR TIMES was ChatGPT's second most cited domain in both April and June 2026, and first for the food industry
- Whether a release gets cited depends on type-specific elements — Sighted (Tech Knowledge Base) ran 1,056 tests through OpenAI's API in May and June 2026. Research-report releases needed five elements in the text: study name, the organization that ran it, sample size, period and percentage figures. Product releases needed product name, launch date, price and location. For product releases, mid-sized companies were cited 78% of the time versus 30% for major chains
- Query wording decides between the press release and the official site — The same study found questions containing "survey," "ranking" or "limited edition" favored PR TIMES, while "recommended," "adoption status" or "new product" favored the official site. You need both in shape
- note, Ameblo, Yahoo! Chiebukuro and Rakuten each play a different role — note and Ameblo hold long-form first-hand accounts, Chiebukuro holds Q&A, Rakuten holds product facts. Stimulate UGC rather than fabricate it, and never lean on a single platform, as the August 2026 citation shift showed
Below, I cover press releases, the release-versus-official-site split, note and Ameblo, then Q&A, marketplaces and reviews, ending with a 30-day plan.
1. How Often ChatGPT Cites Press Releases
First, the scale. In July 2026, Wallabee published an analysis of 81,895 AI answers collected on its AI search monitoring platform Optyino.ai between April 21 and July 8, 2026, covering 120 prompts in finance, housing, appliances and B2B services across nine AI models.
| AI model | Share of answers citing at least one press release |
|---|---|
| ChatGPT (GPT-5 mini) | 42.84% |
| ChatGPT (GPT-5.2) | 27.33% |
| Perplexity | 21.45% |
| Google AI Mode | 20.68% |
| Claude | 6.00% |
| Copilot | 2.90% |
What surprised me is the spread between models. Within ChatGPT alone, GPT-5 mini and GPT-5.2 differ by more than 15 points, and the gap to Copilot is roughly 15x. How much a press release matters depends entirely on which AI is answering.
The way releases get cited matters too. Press releases appeared as source links in 15.05% of answers but were mentioned in the answer text itself in only 0.63%. They're used as footnote-style evidence far more than as named sources in the prose.
And of the 9,381 press release URLs cited, 60.5% were on PR TIMES. That matches Ahrefs' Japan data, where PR TIMES held second place among ChatGPT's cited domains in both the April and June 2026 editions and was first for the food industry (2,445 responses) in the September to December 2025 industry analysis.
For the full picture of citation sources in Japan, see The Domains ChatGPT Cites in Japan.
References: How often does generative AI cite press releases? Optyino.ai analyzes 81,895 AI answers (Japanese) - Wallabee via PR TIMES Press releases appear in about 43% of ChatGPT answers (Japanese) - Manamina Top Japanese domains cited by AI, June 2026 (Japanese) - Ahrefs 2025 AI search trend analysis: industry citation patterns (Japanese) - Ahrefs
2. What Gets a Press Release Cited
Publishing on PR TIMES doesn't guarantee a citation. Sighted, run by Tech Knowledge Base, published on July 15, 2026 the results of seven experiments totaling 1,056 AI search tests through OpenAI's Responses API with web search, analyzing 526 citation passages drawn from releases by more than 70 companies on PR TIMES. The conditions differ by release type.
| Release type | Elements required for citation | Notes |
|---|---|---|
| Research report (5 elements) | Study name in the title; the organization that ran the study; sample size (N=…) stated in the body text; survey period; key findings as percentages | Works best when no more authoritative answer source already exists |
| Product or event (4 elements) | Product or menu name; launch date or availability period in the body; price; location or store information | All four together matter. Mid-sized companies cited 78%, major chains 30% |
| Ranking or roundup (3 elements) | Ranking name and basis; ranks and item names listed as text; categories and conditions separated | Only 2 of 32 were cited. The hardest type |
Write figures and proper nouns as text
The common thread in the research-report type is whether the items the AI extracts are written as text. Study name, the organization behind it, sample size, period and percentages don't count if they only exist in a chart image. Put them in plain prose. We follow the same rule on this blog: any figure that appears in a graphic also appears in the body text.
Mid-sized companies get cited more
A 78% citation rate for mid-sized companies versus 30% for major chains on product releases may seem counterintuitive. My interpretation: major chains already have more authoritative answer sources, official sites and major media coverage, so the AI doesn't need to reach back to the release. It's the same structure as the research-report condition of "no more authoritative source already exists." For mid-sized companies, the press release is the shortest path to being cited as their own primary source.
3. Press Release or Official Site?
The most practical finding in Sighted's study, to me, is that the citation target switches based on the wording of the question. For the same company and the same information, how the user phrases the query determines whether PR TIMES or the official site gets cited.
| Words in the question | Likely citation target |
|---|---|
| "survey," "ranking," "limited edition" | The release on PR TIMES |
| "recommended," "adoption status," "new product" | The official site |
The design that follows is simple. Put surveys, rankings and time-limited information on the press wire; put product recommendation rationale, case studies and evergreen new-product information on the official site, organized heading by heading. Do only one and you miss the questions phrased the other way.
On the official-site side, remember that 16.6% of ChatGPT's sub-queries for brand prompts search the brand's own domain by name. Sections 3 and 4 of How to Get Cited by ChatGPT: The Complete Guide cover the structure.
LinkSurge
linksurge.jp
SEO・AIO・GEO統合分析プラットフォーム。AI Overviews分析、SEO順位計測、GEO引用最適化など、生成AI時代のブランド露出を最大化するための分析ツールを提供しています。
4. note and Ameblo: Where Long-Form First-Hand Accounts Live
In Ahrefs' Japan data, Ameblo and note are consistently near the top of ChatGPT's citation sources. In June 2026, Ameblo was third and note fifth. In April, Ameblo was first. In the industry analysis, Ameblo was third for food, and note, Ameblo and PR TIMES all ranked for IT.
Why they get cited
My view is that these two platforms hold a huge volume of long-form writing by people who actually used, ate or tried the thing, which is exactly the first-hand material ChatGPT leans on as evidence. Of the five fan-out types (related topics, implicit questions, comparisons, recency, rephrasings), the text that answers "implicit questions" and "comparisons" more often lives in individuals' long posts than on official sites. The equivalents elsewhere are Medium, Substack, and community blogs.
How a company uses note
- Publish your own primary information at length — Development history, test data, what went wrong. The depth a press release can't carry belongs on note
- Spell out proper nouns and years — 23.8% of ChatGPT's sub-queries carry exact-match quotes and 19.6% carry a year. Put product names, company name and the study year in the title and body
- Make each heading self-contained — Under every heading, a paragraph answering that heading's question in 40 to 60 words
How to get more third-party note and Ameblo posts
Writing from your own account has limits. To increase posts by people who actually used the product, provide review samples, run hands-on events, and offer templates that make writing easier. Passing off your own writing as a third party's is a platform-policy and trust problem before it's an AI problem. The line between stimulating and fabricating UGC is drawn in UGC and AI Search Citations.
References: Top 10 domains cited by AI search, April 2026 update (Japanese) - Ahrefs A study of query fan-out in ChatGPT using brand searches - MJ Cachón
5. Yahoo! Chiebukuro, Rakuten and Reviews: Q&A and Product Facts
Yahoo! Chiebukuro
Yahoo! Chiebukuro, Japan's largest Q&A site, entered ChatGPT's top 10 at ninth in June 2026. In Google AI Mode it was the overwhelming leader in May to September 2025 with 518,977 citations. Its share in ChatGPT is smaller than in Google, but question-style queries such as "what's the difference between X and Y" or "what to watch out for when choosing X" structurally match Chiebukuro's question titles, which makes it a natural citation candidate.
What a company can do is answer questions in its category accurately, under an official account or a named employee. Answer, don't advertise. Linking to your own primary information (surveys, specs) inside the answer makes both the Chiebukuro answer and your own site citation candidates.
Rakuten, Amazon and reviews
Rakuten marketplace product pages ranked sixth in June 2026 and second for the food industry. Profound's July 2026 analysis found that 87.3% of product information in ChatGPT's shopping answers came from web crawling, and that top-ranked product cards had a median of 787 reviews versus 352 for cards ranked fourth or lower.
For e-commerce, the tactics are accurate specs, pricing and availability as text on marketplace product pages, and a mechanism for growing review counts (post-purchase requests, a clear path to posting). Shopping-specific tactics get their own article in this series.
Price in the volatility risk
In August 2026, Reddit's share of ChatGPT Search citations fell 86% in four days, and YouTube and TikTok citations dropped 70 to 90 percent (cause unconfirmed). UGC platforms matter a lot when they matter, and can vanish overnight. Don't build around one. See ChatGPT's Citation Sources Moved 86% in Four Days.
References: Ahrefs analysis of AI search citation sources in Japan (Japanese) - Manamina Breaking down how ChatGPT Shopping works behind the user experience - Profound Reddit Citations Are Dropping in ChatGPT - Promptwatch
6. A 30-Day Plan for the Three Channels
Here's the sequence.
| Week | Do | Verify with |
|---|---|---|
| 1 | Log ChatGPT's cited URLs for 20 to 30 category questions and group them into five buckets: press wire, note/Ameblo, Q&A, marketplaces, official site | Identify buckets where competitors are cited and you aren't |
| 2 | Publish one research-type or product-type release that meets Sighted's conditions (study name, the organization behind it, N, period, figures; or product name, launch date, price, location) | Log citations for two weeks after publication |
| 3 | Build official-site pages that answer "recommended," "adoption status" and "new product" questions heading by heading. Publish one long-form primary-information post on note | Log whether the official site and note get cited |
| 4 | Answer category questions on the Q&A site accurately. Build a review request path | Make weekly logging of mention rate and cited domains routine |
The logging in weeks 1 and 4 runs automatically if you register watch queries in LinkSurge's GEO monitoring, across five engines including ChatGPT Search. We use the same setup to measure this series.
Frequently Asked Questions
How often does ChatGPT cite press releases?
In Wallabee's analysis of 81,895 AI answers from April 21 to July 8, 2026, the share of answers citing at least one press release was 42.84% for ChatGPT (GPT-5 mini) and 27.33% for GPT-5.2. Perplexity was 21.45%, Claude 6.00% and Copilot 2.90%. 60.5% of the cited press release URLs were on PR TIMES.
Does publishing on PR TIMES get you cited by ChatGPT?
Only under conditions. In Sighted's 1,056-run test in May and June 2026, research-report releases needed five elements stated in the text (study name, the organization that ran it, sample size, survey period and percentage figures), and product releases needed product name, launch date, price and location. For product releases, mid-sized companies were cited 78% of the time versus 30% for major chains, and topics that already have a more authoritative answer source are less likely to be cited.
Does ChatGPT cite the press release or the official site?
It depends on the question's wording. Sighted found that questions containing "survey," "ranking" or "limited edition" favored the PR TIMES release, while "recommended," "adoption status" or "new product" favored the official site. Put surveys, rankings and time-limited information on the wire and product rationale and case studies on the official site to reduce misses.
Can companies use note or Ameblo for ChatGPT optimization?
Yes. In Ahrefs' June 2026 data, Ameblo was ChatGPT's third most cited domain and note fifth, and both rank in the industry analysis. Publishing primary information at length from your own account, spelling out proper nouns and years, and making each heading self-contained turns posts into citation candidates. To grow third-party posts, stimulate (samples, hands-on opportunities, templates) rather than fabricate.
Does answering on a Q&A site help with ChatGPT citations?
Yes. Yahoo! Chiebukuro entered ChatGPT's top 10 cited domains in June 2026 and was Google AI Mode's largest citation source with 518,977 citations in May to September 2025. Question-style queries structurally match Q&A titles. Answering category questions accurately under an official or named account and linking to your own primary information makes both the answer and your site citation candidates.
Conclusion: Don't Just Publish Releases. Publish Them in the Form That Gets Cited
ChatGPT optimization in Japan is built on three channels that Western guides don't cover. Press releases appear in up to 42.84% of ChatGPT answers, and 60% of those are on PR TIMES. But whether a release is cited depends on type-specific elements, and query wording decides between the release and the official site. note and Ameblo hold long-form first-hand accounts, Chiebukuro holds Q&A, and Rakuten holds product facts. Each plays a different role, and each has a counterpart in your own market.
Start by logging who's cited for your category's questions in five buckets, then publish one research-type or product-type release that meets the conditions. And don't lean on a single platform. Spread thinly across the five buckets, official site included. Sources rotate within days.
LinkSurge's GEO monitoring records mentions and cited URLs per watch query across five engines including ChatGPT Search. Use it to track citation changes before and after a release goes out.
