Being mentioned in AI answers and being accessible to agents are different strengths.
Bosch, Miele, Electrolux, Samsung and LG, measured on the same day under one protocol: 77 German questions a household asks before buying, two AI systems (ChatGPT and Perplexity), two runs each, then a door test on every brand site. This is a sample of one market on one day. Per-answer ratings and sources are in the evidence.
rated answers per brand
32 open questions all five brands share, plus 8 with the brand's name, times 2 systems, times two runs. Identical basis for every brand.
best score: Samsung
Lowest: Electrolux with 1.8. Same scorer as the public DAX 40 Agent Success Index.
brands passed all sampled access checks
Samsung and LG passed all tested client identities. Bosch and Miele refused several requests; Electrolux timed out for non-browser requests.
the most any brand is named without being asked for
In the open questions nobody says a brand name. Bosch appears most often; Electrolux least, in 7 percent.
When nobody says a brand name, the AI systems name Bosch in 66 percent of the answers and Miele in 47; Samsung and LG appear in 20 and 13 percent, Electrolux in 7. In the access checks, all tested clients reached the sampled Samsung and LG pages. Bosch and Miele refused several attempts; Electrolux timed out for non-browser clients. A mention rate does not tell you whether an agent can use the website.
Score = 0.20 AI Visibility + 0.70 AI Usability + 0.10 Evidence, on 100, shown on 10. Visibility is what the measured AI systems do with the brand; Usability is how far a text, code, browser and fully automated agent gets on the brand site; Evidence is what the site offers machines to check.
| # | Brand | Score / 10 | AI Visibility | AI Usability | Evidence | Named, open questions | Own site cited, with name |
|---|---|---|---|---|---|---|---|
| 1 | Samsung | 6.2 | 18 | 74 | 70 | 26/128 (20 %) | 24/32 (75 %) |
| 2 | LG | 6.2 | 17 | 74 | 72 | 17/128 (13 %) | 23/32 (72 %) |
| 3 | Bosch | 5.5 | 37 | 61 | 45 | 85/128 (66 %) | 24/32 (75 %) |
| 4 | Miele | 5.3 | 29 | 61 | 42 | 60/128 (47 %) | 22/32 (69 %) |
| 5 | Electrolux | 1.8 | 13 | 19 | 25 | 9/128 (7 %) | 20/32 (62 %) |
V 18 · U 74 · E 70
V 17 · U 74 · E 72
V 37 · U 61 · E 45
V 29 · U 61 · E 42
V 13 · U 19 · E 25
On 2 September we called one product page per brand with six client identities (plain curl, python-requests, GPTBot, ClaudeBot, PerplexityBot and a Chrome user-agent string), then checked the site in a real browser. The bot names are test identifiers, not verified vendor crawlers. No purchase was submitted.
| Brand | Text agent | Code agent | Browser agent | Full automation | What we saw |
|---|---|---|---|---|---|
| Boschbosch-home.com | ◐ | ◐ | ✓ | ◐ | Test requests using GPTBot, ClaudeBot and PerplexityBot user-agent strings returned 403; anonymous scripts get product pages; llms.txt returned HTML and refused the tested AI-bot user-agent strings; own cart plus dealer path. |
| Mielemiele.de | ◐ | ◐ | ✓ | ◐ | Only the plain curl identity gets in; python-requests, a browser user agent without browser fingerprint and the tested AI-bot user-agent strings returned 403; rich Product schema; no llms.txt; own cart plus dealer search. |
| Electroluxelectrolux.de | ✕ | ✕ | ✓ | ✕ | The tested non-browser requests timed out after the TLS handshake; a full browser returned 200; no prices, dealer search only; no washing machines in the German range. |
| Samsungsamsung.com/de | ✓ | ✓ | ✓ | ◐ | All tested client identities returned 200; robots.txt allows GPTBot, ChatGPT-User, OAI-SearchBot, PerplexityBot (ClaudeBot not named); llms.txt in the right format but without Germany; Product and Offer schema on the category page; own shop. |
| LGlg.com/de | ✓ | ✓ | ✓ | ◐ | All tested client identities returned identical bytes; the most explicit AI-bot allow list of the five; llms.txt in the right format but without Germany, and /de/llms.txt is a soft 404; Product and Offer schema; own shop. |
✓ the tested access profile reached the relevant information or path · ◐ partial access · ✕ the tested profile did not reach it. Real provider crawlers may receive different treatment. A refusal or timeout does not establish the website operator's intent, nor explain the AI mention rate.
Two separate question sets: 32 unbranded questions and 8 questions naming each brand. In the unbranded set, mentions and own-site citations use the same denominator of 128 answers. They are independent measures: a source can be cited without the brand name appearing in the answer text. A mention is not necessarily a recommendation.
| Brand | Open: named | Open: own site cited | Open: competitor, no own source | With name: own site cited | Most frequent alternative brands/sources |
|---|---|---|---|---|---|
| Samsung | 20 % | 1 % | 66 % | 75 % | Bosch 80 · Miele 17 · Bauknecht 4 · Electrolux 3 |
| LG | 13 % | 2 % | 72 % | 72 % | Bosch 82 · Miele 17 · Bauknecht 4 · Electrolux 3 |
| Bosch | 66 % | 23 % | 14 % | 75 % | Miele 52 · Samsung 7 · Siemens 6 · Electrolux 5 |
| Miele | 47 % | 16 % | 38 % | 69 % | Bosch 78 · Bauknecht 4 · Electrolux 3 · AEG 2 |
| Electrolux | 7 % | 0 % | 80 % | 62 % | Bosch 84 · Miele 17 · LG 4 · AEG 4 |
All tested client identities could access the sampled product page. Product and Offer data were available on the category page, and a direct shopping path was visible. Samsung was named in 20 percent of the unbranded answers, with its own site cited in 1 percent. Access and visibility are different findings.
Visibility per task: Washing machine 27.81 · Dishwasher 16.56 · Fridge-freezer 20.47 · Oven and kitchen 17.19 · Situation scenarios 6.25
All tested client identities could access the sampled product page. LG was named in 13 percent of the unbranded answers, with its own site cited in 2 percent. Product-page structured data was added by JavaScript in the browser. The global llms.txt omitted Germany; /de/llms.txt returned an error page with HTTP 200.
Visibility per task: Washing machine 17.03 · Dishwasher 16.56 · Fridge-freezer 25.47 · Oven and kitchen 19.06 · Situation scenarios 6.41
Named in 66 percent of the unbranded answers, with its own site cited in 23 percent of the same answers. Other sources also describe the brand. Several test clients using AI-bot user-agent strings received 403 responses, while a browser could access the sampled page.
Visibility per task: Washing machine 51.88 · Dishwasher 48.44 · Fridge-freezer 25.94 · Oven and kitchen 31.56 · Situation scenarios 29.38
Named in 47 percent of the unbranded answers; its own site was cited in 16 percent. For the separate set of questions that explicitly named Miele, its own-source rate was 69 percent. The sampled product page had detailed structured data, but several automated requests were refused.
Visibility per task: Washing machine 30.94 · Dishwasher 31.09 · Fridge-freezer 25.63 · Oven and kitchen 33.28 · Situation scenarios 24.69
Named in 7 percent of the unbranded answers, with no own-site citations. Its German laundry offer sits under AEG, a sister brand counted separately. The sampled page was accessible in a full browser; the tested non-browser requests timed out. This does not establish that every agent is blocked.
Visibility per task: Washing machine 14.53 · Dishwasher 13.91 · Fridge-freezer 16.56 · Oven and kitchen 19.06 · Situation scenarios 2.34
Provide reliable access for customer tasks.
Test access with the agents relevant to your customers. A robots.txt rule alone does not ensure that a request reaches the page. Check server and security behaviour as well as what a full browser can do. Allow legitimate customer tasks through appropriate controls.
Put the answer on the brand page, in server HTML, with structured data.
Provide verified model numbers, prices, compatibility, noise and energy figures in accessible text and structured data. Keep them consistent with the visible product page. A particular file or format alone does not guarantee citation.
Ask the same 77 questions again.
Repeat the frozen question set and access checks after changes. Compare what improved and what remains difficult. Models, search results and other sources also change; a before-and-after difference alone does not prove causation.
Questions. 77 German questions, frozen before the run: four purchase tasks (washing machine, dishwasher, fridge-freezer, oven and kitchen) with 6 open questions each and 2 per brand with the name, plus 8 situation scenarios a household would describe in its own words, plus 5 knowledge controls. Coverage gate against the 13-category question taxonomy: passed.
Systems. ChatGPT and Perplexity, each with web search, two runs per question, over the public APIs with neutral profiles on 2026-09-02. No personal accounts, no browser sessions, no location hints beyond the German question. This published comparison uses ChatGPT and Perplexity only: four answers per question. Separate Claude results are outside this report.
Mentions. Name-in-text rates were recounted from the original answers on 22 September 2026. They count the brand name independently of own-site citations. The earlier display combined those two events. This clarification changes mention counts, not the recorded Agent Success Scores or source counts.
Rating. Host-strict. A brand's own site counts only when its registered host or a true subdomain is among the cited sources. Sister brands of the same group (Siemens, Neff and Gaggenau for Bosch; AEG for Electrolux) are noted but never counted as own. For ChatGPT only sources linked in the answer text count; for Perplexity the cited sources and the search results.
Door test. One product page per brand, six client identities, then a real browser. Four agent classes rated success, partial or blocked; the close state is the deepest point reached without submitting anything.
Controls. 15 of 20 control answers searched the web although the question was pure knowledge. That is a property of the systems, not of any brand.
Limits. One market, one language, one day, one product page per brand for the door. A sample, not a census; a full analysis asks 100 to 200 questions per brand. Electrolux sells laundry in Germany as AEG, so "Electrolux washing machine" is a question about a product the brand does not offer here; we kept it because the comparison was asked for by brand name, and we say so wherever it matters. A named bot in our door test is a user-agent string, not the vendor's crawler.
Brands can report factual errors or submit a response to hello@hyperize.ai. We review corrections against the recorded evidence, re-test where warranted and publish relevant responses alongside the findings.
The Agent Success Score is a snapshot of how public digital touchpoints perform for AI agents at the time of measurement. It makes no statement about product quality, company performance, service quality or legal obligations of the named brands. Brand names remain with their owners and are used solely for identification and reporting within editorial coverage (§ 23 MarkenG, Art. 5 GG).
