Article voting

Every published article has a small voting widget at the bottom asking "Was this helpful?". Visitors click yes or no. That is it. The simplicity is the point. The data drives content decisions and AI retrieval quality.

Where it appears

Below the article body, above the footer. Two buttons: a thumbs up and a thumbs down. After voting, the buttons collapse into a thank-you message.

You can hide the widget per help center on Knowledge → Settings if you do not want to collect votes. We strongly recommend leaving it on.

Who can vote

Anyone reading the article. No sign-in required. Votes are anonymous.

We deduplicate by browser fingerprint plus a session cookie, so the same visitor cannot vote ten times in a row. The dedupe window is 30 days per article. After that, the same visitor can vote again on the same article (their experience may have changed).

Comment box

When a visitor votes "unhelpful", a small free-text comment field appears asking why. The comment is optional. If they fill it in, you will see it in the article's analytics view alongside the vote.

Comments are the single highest-signal feedback you will ever get. Read them.

A few common patterns:

• "I couldn't find X" — your article is missing a section.
• "This is for Y but I have Z" — your title is misleading.
• "I tried this and it didn't work" — your article is wrong or outdated.

Each one is a concrete TODO.

Vote totals

Votes appear as raw counts in the article analytics view. See Article analytics for the full breakdown.

The two metrics that matter:

• Helpful rate. Percentage of votes that were helpful. Anything below 60% is an article that needs work.
• Total votes. Volume tells you confidence. Five unhelpful out of five votes is signal. Five unhelpful out of 500 is noise.

How votes affect AI retrieval

Articles with consistently low helpful rates get downweighted in AI retrieval. The agent will still see them, but they will rank lower against more reliable articles. See The brain: KB graph.

The downweighting is gradual and reversible. Fix the article, votes recover, weight goes back up. There is no permanent penalty.

This is why we encourage you to leave voting on even if you do not act on the data immediately. The AI is acting on it for you.

Customizing the prompt

The default question is "Was this helpful?". You can change it on Knowledge → Settings. Variants we have seen work:

• "Did this answer your question?"
• "Was this article useful?"
• "Found what you needed?"

Do not get cute. The question should match the action.

Vote spam

We deduplicate aggressively and rate-limit by IP. We have seen voting spam essentially never. If you do see a sudden burst of negative votes that looks like bot activity, contact support and we will investigate.

Adjusting votes manually

You cannot manually delete or adjust votes. If an article racked up unhelpful votes during a period when it was wrong, fix the article and the helpful rate will recover from new votes. The history stays so you can see the impact of the fix.

If you really need to reset, you can clone the article, copy the body to a new slug, and archive the old one. New article starts at zero votes. We do not recommend doing this; the historical data is more useful than a clean slate.

Voting and gated help centers

Voting works the same on gated help centers. Visitors are still anonymous to you (we do not tie votes to email-verified identity). The deduplication is per browser session as before.

Voting and the AI agent's drafts

When the AI agent uses an article in a reply, the customer does not see a vote prompt for that article in chat. Voting is help-center-only. We measure AI quality separately through AI receipts and CSAT on AI replies. See CSAT overview.

Disabling voting per article

You cannot disable voting on one article. It is a workspace-wide toggle. If a particular article should not have voting (for example, a "service status" type page), consider whether it should be in the help center at all or somewhere else.

Practical advice

The instinct is to obsess over votes. Do not. The instinct is to ignore votes. Also do not.

The healthy cadence is:

• Once a month, look at the bottom three articles by helpful rate.
• Read the unhelpful comments on each.
• Pick one specific change for each article based on the comments.
• Rewrite. Republish. Move on.

Most articles will not need attention most of the time. The ones that do, will tell you exactly what is wrong if you read the comments.

Related

• Article analytics for the dashboard.
• The brain: KB graph for retrieval weighting.
• Authoring articles for the rewrite mechanics.