PG&E surfaces in AI answers about safety and what to avoid, not reliability or trust

PeakMetrics asked ChatGPT, Claude, Gemini, Perplexity and Grok 250 unbranded energy and utilities questions each on 2026-09-29, for 1,250 answers. PG&E was named in 110 of them. By assistant, Grok named it in 35 of its 250 answers, ChatGPT 25, Perplexity 19, Claude 17 and Gemini 14.

The question type mattered more than the assistant. PG&E's visibility was highest on questions framed around what to avoid (24%), then safety (19%) and employer questions (18%). It was 0% on reliability and 0% on trust. Service was 13% and innovation 12%. Price and value was 2%.

The stored answers show the pattern. Asked which pipeline operators have the worst safety violations, Claude named PG&E first and described the 2010 San Bruno explosion. Asked about the worst customer service, Perplexity cited a 2020 satisfaction ranking that placed PG&E at the bottom of the U.S. providers it compared. Asked which utilities charge the highest rates, Perplexity listed PG&E third at about 38.6¢/kWh. Asked about scandals or dividend cuts, Grok called PG&E "one of the most prominent cases" and walked through the wildfires, the 2019 bankruptcy and the dividend suspension.

The same answers also carried other framings. Claude described PG&E's EV2-A time-of-use rate as a saver for EV owners. ChatGPT and Perplexity listed PG&E among utilities testing new technology, including a vehicle-to-everything and virtual power plant effort and an emerging-tech program. Claude also reported a 2026 ranking that put PG&E as the top utility employer for women.

The credit tags on these appearances were thin and scattered. Three were "Not Substantive", two "Price & Value" and two "Safety & Compliance". Single tags covered sustainability, reliability, labour and customer experience. No one theme dominated.

The PeakMetrics news index recorded 5,506 mentions of PG&E in the last 30 days, 1,257 of them with the company in the headline. The recent headlines center on corporate and policy events. One reports a strategic review and a 10% share drop. Another covers plans to cut capital spending after a wildfire liability bill failed. A third is a PG&E statement on Senate Bill 492.

Two headlines are operational. One covers a home EV charging bundle with GM Energy. The other warns of wind-driven power shutoffs in Northern California.

The press story is current and financial: liability, spending, legislation and outage warnings. The answers drew more on established reference points. These included the San Bruno explosion, the 2017 dividend halt, the 2019 bankruptcy and a 2020 customer satisfaction ranking. The recent news items were not what the assistants surfaced in these stored answers.

Coverage volume also does not translate directly into answer presence. In this snapshot, Duke Energy had similar 30-day news volume (5,446 mentions) and was named in 127 answers. NextEra Energy had 6,749 mentions and was named in 151. Southern Company had far fewer mentions (2,044) and was named in 65. PG&E's 110 sits between these. The answers seem to follow how questions are framed more than how much the press writes.

For a PG&E comms team, the snapshot points to question framing as the variable to track. The company appears when a question invites a warning, a safety comparison or a price ranking. It does not appear when the question is about reliability or trust. The assistants also did offer non-crisis framings, in EV rates, grid technology and employer rankings, so those threads are already in the answers.

Check which sources sit behind the older reference points, such as the 2020 satisfaction ranking and the San Bruno account. Also check whether recent PG&E announcements are reaching the assistants at all.

This is one run of live answers on a single date. The assistants vary between runs, and the 11 stored answers are a sample of the 110 that named PG&E. Read the figures as a snapshot, not a trend.