Home-Appliance Sale & Campaign AI Capture-Rate Survey: How Far Can Five Major Generative AIs Guide Users to Retailers' Official Campaigns?

Summary

Stellagent Inc. investigated how far major generative AIs can guide shoppers to the sale and campaign information posted on home-appliance retailers' official sites. We fixed three campaigns each — nine in total — that were confirmed to be running as of July 29, 2026 on the official sites of Bic Camera, Nojima, and Kojima, and asked ChatGPT, Gemini, Claude, Perplexity, and Copilot under identical conditions. This survey was conducted by Stellagent Inc. with Codex as the operator.

As a result, out of the 90 capture opportunities for the nine official campaigns, AI answers explicitly captured an official fixed campaign in 39 cases, for a capture rate of 43.3%. By AI, ChatGPT and Gemini were at 27.8%, Claude at 50.0%, and Perplexity and Copilot at 55.6%. By brand, Bic Camera was 70.0%, while Nojima and Kojima were each 30.0%. Here, "capture" means whether an official fixed campaign was explicitly named; the accuracy of the deadlines, conditions, and official URLs presented was outside the scope of this tally.

Metric	Result
Target brands	3 (Bic Camera, Nojima, Kojima)
Target AIs	ChatGPT, Gemini, Claude, Perplexity, Copilot
Survey date	July 29, 2026
Official fixed campaigns	9 (3 brands × 3 each)
AI answers	30 answers (3 brands × 5 AIs × 2 prompts)
Official campaign capture opportunities	90 (9 × 5 AIs × 2 prompts)
Official campaigns captured	39 (43.3%)
Official campaigns not captured	51 (56.7%)
Answers capturing all 3	8 / 30 answers
Answers with partial capture	12 / 30 answers
Answers that did not explicitly capture the fixed targets	10 / 30 answers
Answers with an official-URL candidate string in the body text	8 / 30 answers

The chart above may be reproduced as-is in media coverage, articles, and other materials, provided that you credit "Stellagent Inc." as the source and include a link to this page.

The results confirm that even information posted on official sites is not necessarily presented reliably in an AI shopping consultation. The full text of the prompts we actually entered is provided in the appendix "Prompts Used" at the end of this article.

Background

When buying appliances, many conditions bear on the purchase decision: price, stock, delivery, installation, warranty, points, trade-in, and limited-time sales. For large appliances such as air conditioners, refrigerators, washing machines, and TVs in particular, not only the product price but also each retailer's campaigns and region-limited offers affect the decision.

Meanwhile, consumers gather information not only via search engines and price-comparison sites but increasingly via generative AIs such as ChatGPT, Gemini, Claude, Perplexity, and Copilot. When a user asks an AI, "If I buy an appliance this weekend, which campaigns can I use?", it matters whether the AI can find the official campaign information, organize the deadlines and conditions, and guide the user along with the official URL.

This survey does not evaluate each retailer's campaign content or corporate stance. We conducted it to verify how far the promotional information posted on official sites is captured in a general-purpose AI's shopping consultation, and to clarify the information-readiness issues that EC and retail businesses need to address in the AI era.

Survey Overview

Item	Content
Survey name	How far can major generative AIs guide users to home-appliance retailers' latest sale and campaign information?
Conducted by	Stellagent Inc.
Operator	Codex
Method	AI answer experiment; desk research and cross-checking of official sites
Survey date	July 29, 2026
Target brands	Bic Camera, Nojima, Kojima
Target AIs	ChatGPT, Gemini, Claude, Perplexity, Copilot
Selection criteria	Among major brands that are easy to compare with existing home-appliance AI/LLMO surveys and that consumers readily consult when buying appliances, we targeted those for which running campaigns could be fixed on the official site as of July 29, 2026. To meet same-day release preparation, we set as the initial gate the three brands — Bic Camera, Nojima, and Kojima — for which three official campaigns each could be fixed, and treated expansion to all nine brands as an additional survey.
Procedure	From each brand's official site, we fixed three home-appliance / digital-appliance campaigns each that could be confirmed as running as of July 29, 2026. We then entered Prompt A and Prompt B into each AI and saved the full answer, timestamp, model / display name, plan, web-search setting, and screenshots.
Context-influence control	Prompt A and Prompt B were run in separate chats, separate threads, or temporary chats even for the same AI and same brand. We did not pass the campaign names, URLs, or verdicts obtained in Prompt A into Prompt B. At the top of each log we recorded that it was obtained in an independent chat (context_isolation: independent_chat).
Number of runs	We ran each AI × brand × prompt once.
Aggregation unit	The headline metric is the 90 capture opportunities for the nine official fixed campaigns. As a supplementary metric, we tallied "all 3 captured," "partial capture," and "not captured" per answer across the 30 answers. In this initial gate, we tallied as "captured" whether an official fixed campaign was explicitly named, and the accuracy of the presented deadlines, conditions, and official URLs was outside the scope of individual evaluation.
Recording	We saved the official campaign fixation table, AI answer logs, screenshots, the tally sheet, a normalization memo, and a quality-check memo.
Exclusions and re-runs	Capture failures due to UI operation were re-run with the same prompt in a new chat. In the end we obtained all 30 answers and included them in the analysis.
Ethics and privacy	We targeted only public information and AI output; no personal information was handled.

AI Services and Conditions

For each answer, the model name, plan, and web-search setting were recorded into the log based on the AI service's UI display. An AI service's UI and model display may change with the date and time, account state, and service-side updates. The AI services were not aligned to the same model tier or subscription plan.

AI service	Main display name / model shown	Conditions
ChatGPT	ChatGPT	Temporary chat or new chat. Where a web-search button was available, we enabled it.
Gemini	Gemini Flash-Lite	Run in a new chat in the Gemini web UI.
Claude	Claude Opus 5 High	Run in a new chat in the Claude web UI. Web-search execution was shown during the answer.
Perplexity	Perplexity default	Run in a new thread in incognito mode. Answers were obtained in search mode.
Copilot	Copilot Smart	Run in a new chat in the Copilot web UI, with a temporary-mode indicator present.

Because the model tier, subscription plan, and web-search setting differ by service, the by-AI differences described below include not only differences in the service itself but also differences from the combination of model, plan, and search setting used.

Official Fixed Campaigns

Official campaigns were fixed from those posted on brand official EC, official campaign lists, official store information, official flyers, and official feature pages, and confirmed as of July 29, 2026 to be running or not yet ended. The official pages were confirmed on July 29, 2026 and screenshots were saved. Campaigns whose conditions could only be confirmed after login or inside an app, leaving no public URL for third-party verification, were excluded. The full adoption and exclusion criteria are in the appendix "Observation and Exclusion Criteria (Appendix)" at the end of this article.

Brand	Official fixed campaign	Type	Official confirmation URL
Bic Camera	Online SUMMER SALE	Sale	https://www.biccamera.com/bc/c/sale/daily/index.jsp
Bic Camera	Up to 10-year free warranty on designated air conditioners	Campaign	https://www.biccamera.com/bc/c/sale/daily/index.jsp
Bic Camera	Up to 50,000 yen off on trade-in	Trade-in campaign	https://www.biccamera.com/bc/c/sale/daily/index.jsp
Nojima	Nojima flyer	Flyer	https://www.nojima.co.jp/campaign/chirashi
Nojima	Salt-candy giveaway! Special campaign for heatstroke prevention and the 2027 air-conditioner issue	Campaign	https://www.nojima.co.jp/campaign/ac2027
Nojima	Nojima 67th-anniversary thanksgiving! No-loss! Win up to 10,000 d-points campaign	Point campaign	https://www.nojima.co.jp/campaign/dpoint/
Kojima	Summer stock-clearance sell-off	Coupon / sale	https://www.kojima.net/ec/prod_list.html?CMPN_PLAN_CD=10019614&keyword=%E3%82%AF%E3%83%BC%E3%83%9D%E3%83%B3
Kojima	Same-maker air-conditioner bulk-buy campaign	Bulk-buy campaign	https://www.kojima.net/ec/Special/campaign/ac/multiple/index.html
Kojima	Review-posting campaign, July 2026	Review campaign	https://www.kojima.net/ec/Special/campaign/review.html

Verdict Criteria

Each answer was judged as captured, partially captured, or not captured based on how many of the brand's three official fixed campaigns it explicitly named. The state of official-URL presentation was recorded in three categories based on the presence of the string in the body-text log.

Category	Definition
Captured	An answer that named a campaign identical or substantially identical to an official campaign. At the answer level, we counted "captured" when all three of the brand's official fixed campaigns were explicitly named.
Partial capture	An answer that explicitly named one or two of the brand's three official fixed campaigns.
Not captured	An answer that did not explicitly name the brand's three official fixed campaigns.
Official-URL candidate present	An answer whose body-text log contained an official-URL candidate string that included the brand's official domain along with a path.
Domain string only	An answer whose body-text log showed the brand's official domain name but not an official-URL candidate string including a path.
No link or in-UI citation only	An answer whose body-text log did not even contain the official domain string. Link cards and citation labels in the AI UI were treated as supplementary evidence in the screenshots, separate from the URL strings in the body-text log.

This survey does not judge that "a campaign does not exist." We tallied only whether the official campaigns we fixed this time appeared explicitly in the AI answers.

Scenarios Used

We asked each AI two scenarios: brand-by-brand campaign confirmation (Prompt A) and a consultation in a purchase context (Prompt B). The full text of the prompts is in the appendix "Prompts Used" at the end of this article.

ID	Scenario	Condition
Prompt A	Brand-by-brand campaign confirmation	Specify a brand name and ask, in table form, for the home-appliance / digital-appliance sales or campaigns running on the official site or official online store
Prompt B	Purchase-scenario type	In the context of planning to buy an air conditioner, refrigerator, washing machine, or TV this weekend, ask which official campaigns might be usable

Prompt A was set to see whether the AI could broadly pick up the brand's official campaign list. Prompt B was set to see, in a context close to an actual purchase consultation, whether the AI could guide the user — with deadlines and conditions — to campaigns usable for buying a large appliance.

Result: The Official Campaign Capture Rate Was 43.3%

For the nine official fixed campaigns, five AIs × two prompts created 90 capture opportunities in total. Of these, AI answers explicitly captured an official fixed campaign in 39 cases, for a capture rate of 43.3%.

Metric	Count	Share
Official campaign captured	39 / 90	43.3%
Official campaign not captured	51 / 90	56.7%

At the answer level, 8 of 30 answers captured all three of the brand's official fixed campaigns. Partial capture was 12 of 30 answers, and answers that did not explicitly capture the fixed targets were 10 of 30.

Answer-level verdict	Answers	Share
Captured all 3 official fixed campaigns	8 / 30	26.7%
Partial capture	12 / 30	40.0%
Did not explicitly capture the fixed targets	10 / 30	33.3%

Result: By-AI Capture Rates Ranged from 27.8% to 55.6%

By-AI official campaign capture rates were 27.8% for ChatGPT and Gemini, 50.0% for Claude, and 55.6% for Perplexity and Copilot. Even AIs with strong search and citation features did not stably pick up every campaign we fixed this time.

AI service	Captured	Capture rate
ChatGPT	5 / 18	27.8%
Gemini	5 / 18	27.8%
Claude	9 / 18	50.0%
Perplexity	10 / 18	55.6%
Copilot	10 / 18	55.6%

This difference may have been affected by each service's search feature, citation display, how it picks up official sites, and its caution when answering. However, this is an exploratory survey with one run per AI × brand × prompt, and does not assert a permanent ranking of the AI services.

Result: By Brand, Bic Camera Was 70.0% and Nojima and Kojima 30.0%

By-brand official campaign capture rates were 70.0% for Bic Camera and 30.0% each for Nojima and Kojima.

Brand	Captured	Capture rate
Bic Camera	21 / 30	70.0%
Nojima	9 / 30	30.0%
Kojima	9 / 30	30.0%

The chart above may be reproduced as-is in media coverage, articles, and other materials, provided that you credit "Stellagent Inc." as the source and include a link to this page.

For Bic Camera, the Online SUMMER SALE, the up-to-10-year free warranty on designated air conditioners, and the up-to-50,000-yen-off trade-in were captured by multiple AIs. On the other hand, in Prompt B some answers stated that "no nationwide common campaign for large appliances can be confirmed," so in the purchase scenario the presentation of official fixed campaigns sometimes dropped.

For Nojima, capture of the flyer, the 2027 air-conditioner-issue campaign, and the 67th-anniversary d-point campaign — all posted in the official campaign list — was unstable. In particular, some AIs did not explicitly name the three official fixed campaigns in Prompt A.

For Kojima, the official campaign list included seasonal-appliance sales, air-conditioner bulk-buy, review posting, payment-provider campaigns, and maker campaigns. AI answers, too, tended to mix running offers, ended offers, and third-party offers, so the capture rate for the three fixed campaigns stayed at 30.0%.

Result: 8 of 30 Answers Had an Official-URL Candidate String in the Body Text

In this survey, we asked each AI to show official URLs. Of the 30 answers, 8 had an official-URL candidate string including a path in the body-text log. 3 answers showed only the official domain name, and 19 answers did not even contain the official domain string; combining the latter two categories, we treat 22 answers as "answers without an official-URL candidate string including a path in the body text."

Metric	Answers	Share
Official-URL candidate string with a path confirmed in the body text	8 / 30	26.7%
Domain string only (3), or no link / in-UI citation only (19)	22 / 30	73.3%

However, this figure is affected by the display specifications of each AI UI. In ChatGPT, Claude, Perplexity, and Gemini, official pages were sometimes shown as link cards, citation labels, or source displays within the answer, but in some cases the actual URL string did not remain in the body-text log. Therefore this metric is a supplementary indicator of whether an official URL usable for a purchase decision was made explicit as body text, and does not directly show whether the AI internally accessed the official page.

Even so, for users it is important to be able to check not only the campaign name but also the deadline, conditions, target categories, and official URL together. When the official URL is not made explicit, the user must either trust the AI answer as-is or search separately to re-confirm it.

Main Patterns of Misinformation and Non-capture

Answering "cannot be confirmed" even though the official campaign exists

For Bic Camera's Prompt B, while we had confirmed the Online SUMMER SALE, the up-to-10-year free air-conditioner warranty, and the up-to-50,000-yen-off trade-in as official fixed campaigns, some AI answers concluded that "no nationwide common campaign for large appliances can be confirmed." This is likely a case where the AI leaned too cautiously at the stage of re-organizing the sale information on the official site into the purchase context of "nationwide common," "for large appliances," and "usable this weekend."

Picking a different offer than the fixed target from among multiple offers in the official list

For Nojima and Kojima, many offers are posted in the official campaign list. AI answers sometimes prioritized Epos Card, municipal measures, payment-provider campaigns, maker campaigns, or sales in a different category rather than the campaigns we fixed as targets. Even for information posted in the official list, which offers the AI judged "useful for a shopping consultation" was not stable.

Mixing in ended or third-party offers

For Kojima, some answers organized the information including payment campaigns, maker-sponsored campaigns, and ended offers. Some answers noted an offer as ended, but in a purchase-consultation context, merely mixing in ended or third-party offers imposes a burden on the user to re-confirm whether an offer is usable.

Method of Extraction, Aggregation, and Normalization

The headline metric is the 90 capture opportunities for the nine official fixed campaigns. For the nine campaigns, five AIs × two prompts produce 10 capture opportunities per campaign, for 90 opportunities in total. For each opportunity, we tallied as "captured" whether that campaign name was explicitly stated in the AI answer. As a supplementary metric, we classified the 30 answers into "captured all 3 official fixed campaigns," "partial capture," and "not captured."

Brand names and AI service names were normalized for spelling variation. Brand names were unified to canonical forms such as Yamada Denki for "Yamada Denki / Yamada" and Yodobashi Camera for "Yodobashi," and AI service names to canonical forms such as ChatGPT for "GPT / OpenAI" and Copilot for "Bing Chat / Microsoft Copilot." Campaigns, sales, fairs, features, coupons, and point rewards were eligible for adoption when the purchase conditions, period, and benefits could be confirmed on the official page. Maker-sponsored campaigns and store flyers were eligible only when posting could be confirmed on the retailer's official site.

This survey is designed to measure capture (present / absent) of whether an official fixed campaign was explicitly named, and does not perform any ranking or score ordering. Therefore there is no situation in which ties need to be broken. Capture failures due to UI operation were re-run with the same prompt in a new chat, and the failed answers were excluded from the tally and re-obtained. In the end we obtained all 30 answers and included them in the analysis.

Implications for Businesses

First, campaign information needs not only to be "posted" but to be arranged in a form that an AI can find, understand the conditions of, and re-present in a purchase context. When the sale name, target category, period, benefit, exclusion conditions, and official URL are scattered across pages, the AI may pick up only part, or prioritize a different offer.

On campaign-list pages, a structure that lets running, ended, region-limited, app-limited, maker-sponsored, and payment-provider-sponsored offers be distinguished mechanically is also important. What reads to a human as a "campaign list" can be hard for an AI to judge — which is a currently usable official campaign and which is reference information.

Assuming AI answers, the explicitness of official URLs becomes more important. If a user cannot get back to the official page from the AI answer, they cannot confirm the deadline, target products, application conditions, region conditions, or app conditions. Campaign pages should have titles, body text, dates, structured conditions, and stable URLs that are easy for an AI to cite.

On top of that, retailers need to regularly audit how their important campaigns are guided by the major AIs. By checking AI answers at moments when purchasing behavior is likely — campaign launch, just before the end, and before the weekend — they can catch non-capture, the mixing-in of ended information, and missing official URLs early.

Notes on the Survey

• This is an exploratory survey based on AI answers and official-site information as of July 29, 2026, and is not intended for statistical generalization.
• The target brands are limited to three as an initial gate. This does not represent the trend of home-appliance retailers as a whole.
• Capture in this survey is judged by whether an official fixed campaign was explicitly named; the accuracy of the presented deadlines, conditions, and official URLs was outside the scope of this tally.
• AI answers may vary with the date and time, model, search setting, conversation history, account state, region, and UI changes.
• The number of trials per AI × brand × prompt is one. Answers may fluctuate even under the same conditions.
• Official campaign information changes over short periods. This survey used the official information that could be confirmed as of July 29, 2026 as its baseline.
• The rate of official-URL candidate strings may be underestimated when link cards or citation labels in the AI UI do not remain in the body-text log.
• This survey does not evaluate each brand's campaign content, price competitiveness, physical-store operations, or customer-service quality.
• The product and service names in this report are trademarks or registered trademarks of their respective companies.

Update History

• July 30, 2026: First edition published

Prompts Used

The [brand name] in each prompt was replaced with the official name of Bic Camera, Nojima, or Kojima before input. The prompts were entered in Japanese (English translations are shown here). Prompt A and Prompt B were run in separate chats or separate threads even for the same AI and same brand, and the Prompt A answer was not passed into Prompt B.

Prompt A: Brand-by-brand campaign confirmation

As of July 29, 2026, please tell me the home-appliance / digital-appliance sales or campaigns running on the official site or official online store of [brand name].

Conditions:

• Prioritize what can be confirmed on the official site.
• Summarize the campaign name, target products or categories, deadline, main benefits, and official URL in a table.
• If something may be ended or unofficial, please note that.

Prompt B: Purchase-scenario type

I plan to buy an air conditioner, refrigerator, washing machine, or TV at a home-appliance retailer this weekend. As of July 29, 2026, please tell me the official sales or campaigns of [brand name] that I might be able to use, with deadlines and conditions.

Conditions:

• Always include the official URL.
• If there are region-limited, app-limited, maker-sponsored, online-only, or similar conditions, separate them.
• If the information cannot be confirmed, do not guess — answer "cannot be confirmed."

Observation and Exclusion Criteria (Appendix)

The adoption criteria for official sources and the exclusion targets are as follows.

Adoption criteria for official sources

• Posted on brand official EC, official campaign lists, official store information, official flyers, or official feature pages.
• Running as of July 29, 2026, or already started and not yet past its end date.
• Related to purchase decisions for appliances, digital appliances, home appliances, air-conditioning appliances, TVs, PCs, and the like.

Exclusion targets

• Third-party media, affiliates, and unofficial coupon roundups.
• Those found only on maker sites, with no posting confirmable on the retailer's official site.
• Those whose conditions could only be confirmed after login or inside an app, leaving no public URL for third-party verification.
• Campaigns already ended as of July 29, 2026. However, when an AI presented ended information, it was recorded as misinformation on the AI-answer side.