During the Microsoft Outlook outage on November 12, 2026, AI engines gave very different answers. In a 5WPR analysis using Profound's tools of 288 answers across four engines, Perplexity named Microsoft in 79.2% of answers, ChatGPT and Gemini each in 65.3%, and a Claude API model in 1.4%. Microsoft's own domains were cited in 19.9% of answers with retrieval evidence.
Last updated: November 12, 2026. Incident details reflect reporting from the afternoon of November 12 and may have changed since.
What are the key findings?
The study found that brand visibility during the outage depended heavily on which AI engine answered the question. The five findings below summarize the data.
- Microsoft was named in 152 of 288 answers (52.8%), and in 143 of the 204 answers to the 17 Microsoft-specific prompts (70.1%, calculated from the per-prompt results).
- Perplexity named Microsoft in 57 of 72 answers, ChatGPT and Gemini named it in 47 each, and the Claude model named it once.
- Apple appeared in 69 answers (24.0%) and Zoho in 54 (18.8%), mostly through ChatGPT and Gemini.
- StatusGator was the most cited publisher, appearing in 44 of 287 answers with retrieval evidence (15.3%), ahead of AP News at 36 (12.5%).
- Microsoft's own domains were cited in 57 of those 287 answers (19.9%).
In this article
What happened during the Outlook outage?
On November 12, 2026, Microsoft said it was investigating reports of users having trouble accessing Exchange Online mailboxes through multiple connection methods in the Americas. The company pointed administrators to incident EX1550543 in the Microsoft 365 admin center, according to an Associated Press report carried by WOSU.
ABC News reported that Microsoft first pointed to infrastructure under elevated load across multiple regions. It later said the issue could affect Microsoft 365 connectivity for users in any region.
For communications teams, the outage raised a second question. A customer who asks an AI engine what is happening hears an answer that the company did not write, and that answer can name competitors, monitoring services or news outlets. The 5WPR study measured what four engines said.
How far did the four AI engines disagree?
The four engines disagreed sharply on whether to name Microsoft at all. Perplexity named the company in 57 of its 72 answers, ChatGPT and Gemini in 47 each, and the Claude model in one.
| Brand | Overall | Perplexity | ChatGPT | Gemini | Claude (API) |
|---|
| Microsoft | 52.8% | 79.2% | 65.3% | 65.3% | 1.4% |
| Apple | 24.0% | 11.1% | 38.9% | 43.1% | 2.8% |
| Zoho | 18.8% | 11.1% | 19.4% | 41.7% | 2.8% |
| Downdetector | 14.6% | 37.5% | 11.1% | 9.7% | 0% |
Source: 5WPR analysis using Profound's tools. Overall rates use 288 answers, and each engine returned 72.
The Claude result needs context. The engine in this study was Anthropic's claude-sonnet-4 model, called through the API with Anthropic's web search tool, so it does not describe the Claude chat product. All 72 of its answers carried web retrieval evidence, yet it named Microsoft once. The study counted recorded brand mentions and did not read the answer text, so it cannot say whether the model answered with product names such as Outlook and Exchange instead of the parent company.
The prompt mix also affects the headline figure. Seventeen of the 24 prompts asked directly about Microsoft service status, reliability, troubleshooting or news. Microsoft was named in 143 of those 204 answers (70.1%). The other seven prompts asked about outage tracking tools in general, and Microsoft appeared in only 9 of those 84 answers (10.7%). The 52.8% overall figure blends the two groups.
Which brands appeared besides Microsoft?
Apple, Zoho and Downdetector were the next most frequently named brands, and each one leaned on different engines. Gemini and ChatGPT named Apple and Zoho far more often than Perplexity did, while Perplexity named Downdetector in 37.5% of its answers.
Microsoft appeared in nine of the 12 answers to "Is Outlook down right now?" Apple appeared in seven of those same 12. Across the full sample, Gemini named Zoho in 30 answers and Fastmail in 20. Those figures count mentions of any kind, including comparisons, references and suggested alternatives. The study did not check why Apple was named, which could reflect Apple Mail or iOS troubleshooting as easily as a competing service.
Gemini also surfaced a detail the other three engines left out. It included Microsoft's incident identifier, EX1550543, in 20 of its 72 answers (27.8%). Administrators who need the technical record would find that identifier useful, and it appeared on one engine only.
Where did the AI answers get their information?
The engines drew on a mix of status monitors, news outlets, reference sites and community forums. StatusGator was the most cited individual publisher, followed by AP News and Variety.
| Source domain | Answers citing it | Share of 287 answers |
|---|
| statusgator.com | 44 | 15.3% |
| apnews.com | 36 | 12.5% |
| variety.com | 35 | 12.2% |
| en.wikipedia.org | 34 | 11.8% |
| mashable.com | 34 | 11.8% |
| reddit.com | 32 | 11.1% |
| learn.microsoft.com | 31 | 10.8% |
Source: 5WPR analysis using Profound's tools. Each domain counts once per answer, and an answer could cite several domains.
Microsoft Learn appeared in 31 answers. Across all Microsoft-owned domains in the study, 57 answers cited at least one company source. That is a fifth of the answers with retrieval evidence, so roughly four in five answers about Microsoft's own outage did not cite a Microsoft page. Yahoo (8.0%) and UptimeRobot (12.2%) also earned owned-domain citations, while Zoom, Slack and Downdetector earned few or none.
What should communications teams do during an outage?
Communications teams should publish one timestamped update that serves both customers and administrators, because AI answers draw on whichever source states the details most clearly. These steps follow from the source mix above. They are 5WPR's suggestions based on the data, and the study did not test whether they change what engines say.
- State which services are affected and which are not.
- Tell users what they can do right now, such as which workaround or alternate access path to use.
- Give the time of the next update, so a monitoring service or reporter can show when the information was last confirmed.
- Name the incident identifier and where administrators can find the technical record.
- Send reporters and status monitors the same timestamped details that appear on the company page, so each source can be checked against the others as the incident develops.
A customer asking when Outlook will return and an IT administrator asking for an incident ID need different levels of detail. A single public update that covers both gives every source the same facts to repeat. Teams that plan for this can also read how crisis communications practice is changing as AI answers become a first stop for customers.
How was the study run?
On November 12, 5WPR ran 24 prompts three times each on four engines: Perplexity (sonar-pro), ChatGPT, Gemini and Claude (claude-sonnet-4 through the API). That produced 288 answers, and all 288 were returned. Of those, 287 included retrieval evidence such as search annotations or grounding records.
Seventeen prompts covered Microsoft service status, reliability, troubleshooting and news. Seven covered broader outage tracking and monitoring tools. Brand mentions used all 288 answers as the denominator, and source citation rates used the 287 answers with retrieval evidence. Appearance rates carry a 95% Wilson interval, and where two brands' intervals overlap the study does not claim one is more visible than the other.
Limits of these findings
The study is a one-day snapshot of these prompts and engine configurations. It combines direct Microsoft outage questions with broader monitoring queries. It did not grade answer accuracy, read answer text for product names, or measure how often the public asked each question. Each engine uses a different model, retrieval tool and answer format, so a gap between engines can reflect system design as much as brand visibility. Forty further names picked up in the answers could not be resolved to a brand and carry no metrics. Related reading is available in the Everything-PR research section and in coverage of generative engine optimization.
Research disclosure: 5WPR conducted the outage study using Profound's tools. Everything-PR News LLC is owned by a shared interest group that includes Ronn Torossian, who founded and chairs 5WPR.
