Claude and the 2026 US midterms: what Anthropic changed

Anthropic says people rarely ask Claude who to vote for, and that its newest models are the most even-handed it has built. In an Oct 14, 2026 post, it also announced official voting information inside Claude.ai and live results on election night. All of it is Anthropic's own account.

This site is independent and is not affiliated with Anthropic. Everything here comes from Anthropic's midterms post, which we reopened on 2026-10-14. We did not test how Claude answers election questions, and we cannot verify Anthropic's figures.

What people actually ask Claude about politics

Anthropic sampled real conversations to see what US users bring to Claude.ai. The post says: "We took a random sample of 250,000 US Claude.ai consumer conversations from September 21 to 27, 2026". These are the shares it reports:

• Politics, public policy or social issues: "about 5.75% of all conversations in the United States on Claude.ai"
• Elections or voting specifically: "About 0.51% of all conversations are specifically about elections or voting."
• Asking who or what to vote for: about 0.01%. Anthropic's reading is that "people rarely ask Claude whom or what to vote for."
• Breaking its election integrity policies: "about 0.002% of conversations".

The same post describes people using Claude to "follow the news, conduct academic or professional research on public policy, learn about candidates, and navigate government services."

Three changes for voters

Anthropic lists what voters will see on Claude.ai in the US:

• A TurboVote banner stays. When people ask about voter registration, polling places or election dates, "we'll continue to display a banner linking to TurboVote". The post calls TurboVote "a nonpartisan resource from Democracy Works".
• Official voting information, soon. The post says: "Later this month, Claude will start drawing official voting information from Democracy Works into its responses on Claude.ai for users in the US." Its example is someone asking where to vote or what ID to bring.
• Results on election night. "Claude will begin showing up-to-date results on election night and in the days that follow"

As of 2026-10-14, the second change was announced for later in the month. We have not seen it live.

The even-handedness claim, and who checked it

Anthropic writes that "Claude Opus 5.5, Sonnet 5.5, and Haiku 5.5 are our most evenhanded to date, scoring 99% on our political evenhandedness evaluation". That is Anthropic grading its own models. The three models have their own pages here: Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5.

The post also names three outside reviewers:

• The Future of Free Speech, "an independent think tank at Vanderbilt University", tested Claude on hundreds of contentious requests.
• The Foundation for American Innovation compared how Claude responded to matched prompts from the left and right.
• The Collective Intelligence Project surveyed more than 2,000 people in the US, UK and India.

What the reviewers found wrong

This part matters more than the 99%. Anthropic says the Future of Free Speech and the Collective Intelligence Project "both found that Claude rarely refused requests, but sometimes added unsolicited caveats and counterarguments, such as inserting an opposing view into an advocacy email."

The Foundation for American Innovation found that "Claude Sonnet 4.5 didn't always hedge evenly, and was more likely to soften arguments for right-leaning positions." That test used an older model than the three named above. We did not rerun it.

Misuse: what Anthropic says it banned

Anthropic describes foreign operators using Claude for influence work. In one case, "foreign-based operators of a fake news site used Claude to produce US midterm coverage favoring Russian, Iranian, and North Korean narratives."

Its response, in its words: "We banned the accounts involved in these cases and improved our ability to detect this type of abuse." The rules behind that are in Anthropic's usage policy, which prohibits generating "false or misleading information in political and electoral contexts".

How to use Claude for election questions

These are our suggestions, not Anthropic's, and we did not run them for this brief. Claude cannot see your ballot, so treat what it says about your local races as a lead to check.

1. Ask for each side, then compare. "Summarize the strongest case for and against this ballot measure, in the words its supporters and opponents would use."
2. Turn on web search for dates and rules. Our guide to using Claude shows where the switch is. Then ask for the source link beside each date.
3. Ask Claude to flag its own slant. "Reread your answer. Where did you hedge more for one side?"
4. Confirm anything that affects your vote (registration deadline, ID, polling place) with your state or local election office.

Step 3 is a response to the reviewers' finding about uneven hedging. It is a prompt, not a fix.

Limits of this brief

• It rests on one source: Anthropic describing its own work. The 99% score and the usage shares come from Anthropic's own tests and sample.
• The post does not say what Claude does when someone asks who to vote for, beyond how rarely that happens. We did not test it.
• We have not seen the Democracy Works integration working. The post places it later in October.
• The numbers cover US consumer conversations on Claude.ai, not the API or other countries.
• No worked example is included. The prompts above are untested suggestions.

If a Claude answer ever looks lopsided, our what is Claude explainer covers how to check it before you rely on it.

Questions

Does Claude tell you who to vote for?

Anthropic's post does not say what Claude does when asked. It reports that about 0.01% of sampled US Claude.ai conversations asked who or what to vote for. We did not test it, so try your own question and judge the answer.

Are the 2026 midterm changes live in Claude today?

Partly. The TurboVote banner continues. Anthropic says official voting information from Democracy Works arrives later this month, and results will show on election night and the days after. We have not seen the new features working.

How even-handed is Claude on politics?

Anthropic scores Opus 5.5, Sonnet 5.5 and Haiku 5.5 at 99% on its own political even-handedness evaluation. Outside reviewers found Claude sometimes added unsolicited caveats and did not always hedge evenly. Treat the score as a vendor claim.

Sources (checked 2026-10-14)

• Anthropic: Preparing for the 2026 US midterms
• Anthropic: Usage Policy