---
title: "Anthropic (Claude) API key"
description: "Add your Anthropic API key to score mentions with Claude: where to get the key, model options and costs, and troubleshooting."
source: https://backtalk.sh/docs/anthropic
---
# Anthropic (Claude) key

Bring your own [Anthropic](https://www.anthropic.com/) key to score and categorise mentions with **Claude**. Without an AI key, BackTalk falls back to a free offline word-list scorer that only labels obvious positive/negative and returns *unknown* a lot; a Claude key gives you nuanced sentiment, categories, and priority on every new signal.

## What it unlocks

- AI sentiment (positive / neutral / negative) instead of the offline lexicon.
- Category and priority tagging on each mention.
- A one-time background **re-score** of your existing mentions when you save the key.

You can save **a key per provider**. Left untouched, your first key powers filtering (sentiment, relevance, priority) and every saved key's model joins the AI share-of-voice comparison. Point either job at any saved key and model, including several models from one provider, under **What runs each task** in Settings.

## Get your API key

1. Sign in at the [Anthropic Console](https://console.anthropic.com/).
2. Add a payment method under **Billing** (API usage is prepaid/pay-as-you-go and separate from a Claude.ai subscription).
3. Open [**Settings → API keys**](https://console.anthropic.com/settings/keys) and choose **Create key**.
4. Copy the key. It starts with `sk-ant-…` and is shown only once.

## Add it in BackTalk

1. Go to [**Settings**](/app/settings/ai) and find the **Claude (Anthropic)** card under AI providers.
2. Pick a model (see below) and paste your `sk-ant-…` key.
3. **Save.** BackTalk sends one tiny test request to confirm the key and model work, then stores the key encrypted and starts the background re-score.

## Models & cost

You're billed by Anthropic directly, at their token rates; BackTalk never marks anything up and only shows an estimate. Sentiment prompts are short, so real spend is typically a fraction of a cent per mention. List prices (USD per million input / output tokens):

- **Claude Haiku 4.5**: ~$1 / $5. Fastest and cheapest; the right default for sentiment.
- **Claude Sonnet 4.6**: ~$3 / $15. More nuance on ambiguous mentions.
- **Claude Opus 4.8**: ~$5 / $25. Most capable; rarely needed just for scoring.

Settings shows a live usage ledger (tokens and estimated cost) once scoring runs.

## Troubleshooting

- **"The provider rejected that API key."** The key is wrong, revoked, or from a different org. Regenerate it in the console and re-paste.
- **"That model isn't available on this key."** Your account can't reach that model yet. Add billing, or pick Haiku.
- **Key check failed / timeout.** Usually no billing on the Anthropic account, or a transient network issue. Confirm billing, then retry.

> **Your key stays private.** It's encrypted at rest and only decrypted server-side at the moment of a scoring call. The plaintext is never sent back to your browser and never written to logs. Settings only ever shows the last few characters as a hint.
>
> **Rotating your key:** create a new key in the Anthropic Console, paste it into Settings to replace the old one, then revoke the previous key.