AI visibility · what it costs to run
AI answer tracking cost: your own keys, the estimate first, a daily cap
AI answer tracking cost is paid to the AI providers, on your own accounts, never through Skald. Each tracking profile shows its estimate before it runs, and stops at its daily cap.
AI answer tracking cost is what the AI providers charge for the answers Skald asks for: the model's tokens (the pieces of words a provider bills for, read and written) plus its web search, for each question, try, assistant and run. Skald works it out before anything runs, and a daily cap keeps the spend there.
30-day free trial · no card · we set it up within one working day · free to self-host
SKALD · AI ANSWERS
New York profile · 5 questions × 3 tries
- ChatGPT15 × $0.03$0.45
- Perplexity15 × $0.01$0.15
- Gemini15 × $0.04$0.60
- Claude15 × $0.03$0.45
- a run60 answers$1.65
- × 4.3 runs a month$7.09
- daily cap$1.65 ≤ $2.00fits
- paid to OpenAI, Perplexity, Google and Anthropic, on your own keys. Your bill comes from each of them.
change how often Sample profile at Skald's default prices an answer. Your server's admin can set them to match your contract.
[01] the estimate, before you save
"How much will this cost?" should not wait for the bill. Set it up and see.
Skald works out a profile's cost before it ever asks: questions × tries × each assistant's price an answer, then × runs a month. Only assistants with a key count. Move the sliders: this is the same sum the profile panel shows. Part of AI visibility tracking.
move a slider The cap goes up to $100 in Skald; the slider stops at $10.
- answers a run
- 60
- a run
- $1.65
- a month, estimated
- $7.09
A run's $1.65 fits under the $2.00 cap: it finishes the day it starts.
5 × 3 × ($0.03 + $0.01 + $0.04 + $0.03) = $1.65 a run · × 4.3 runs = $7.09
30 days free. The answers run on your own API keys, so you see their cost before the first one.
[02] the cap, ask by ask
A cap checked at midnight is a bill you already have. This one is checked before every ask.
Before each answer is asked, its estimated cost is added to what the profile has booked today. If the total would pass the cap, that assistant stops for the day and the run goes on tomorrow. Lower the cap and watch the 60 asks of one run spill into the next days.
Sat 24 · day 1 $1.00 of $1.00
37 asks · the last cent went to Perplexity, the cheapest
Sun 25 · day 2 $0.65 of $1.00
23 asks · the run is done
ChatGPT $0.03Perplexity $0.01Gemini $0.04Claude $0.03failed, still booked
One run, 60 asks, $1.65 at a $1.00 cap: it takes 2 days. Each assistant asks in its own thread, so when the dearest one stops, a cheaper one can still fit.
Sample run: New York, 5 questions × 3 tries × 4 assistants, shown asking in turn. In Skald each assistant asks in its own thread, so the real order varies. A failed ask is booked at its estimate too, and is not asked again in that run.
[03] when a provider says no
A busy API should not lose a run, or hammer the provider. See how long Skald waits.
Every answer has a status code: 200 means here it is, 429 means slow down, 503 means the provider is busy, 401 means the key was refused. Rate limits, timeouts and server errors are tried again up to three times, waiting 4, then 8, then 16 seconds, or longer when the provider asks for it (up to 60). A wrong key or a bad request is not retried: asking again would only fail again.
0 stry 1 → 503 · wait 4 s4 stry 2 → 503 · wait 8 s12 stry 3 → 200 · answer stored
Answered on the third try. One ask, one $0.04 booked: the retries are not booked again.
Waits are between tries, plus the time each request takes. A request gives up after 90 seconds.
[04] your keys, your bill
A key you can read back is a key someone else can read. Put one in and try.
Each assistant is asked on your own account, with your own key: Skald never sees the bill and never resells an answer. An install admin sets each key once. From then on the page only says "set": the key is never shown, printed, logged, or sent anywhere but to its own provider.
OpenAI API key (ChatGPT answers): no secret
GET ai-answers → "keys": {
"openai": false, "perplexity": false,
"gemini": false, "anthropic": false
}
Environment name: SKALD_OPENAI_API_KEY. When it is set, it wins over the saved key.
- 0600 Saved keys live in secrets.json, readable by the server's own user only. A file others can read is refused, not used.
- *** If a provider's error ever repeats the key, it is stored as *** in the failed try's reason.
- $0 No key and no profile: nothing is asked and nothing costs. An assistant without a key is left out, and the page says so.
[05] the real bill
Where to check what AI answers really cost you
On each provider's own usage page. Skald's figure is an estimate made before the run, from default prices of 1 to 4 US cents an answer: the model's tokens plus the web search each provider bills.
| Assistant | Default model | US$ |
|---|---|---|
| ChatGPT | gpt-4.1-mini, web search | 0.03 |
| Perplexity | preset fast, web search | 0.01 |
| Gemini | gemini-2.5-flash, Google Search | 0.04 |
| Claude | claude-haiku-4-5, web search | 0.03 |
What it does not do
- The estimate is a guide. Providers price tokens and web search their own way, and change them.
- Answers run on your keys, at your cost. Skald never pays for them or resells them.
- More answers still give a labelled sample, a direction over weeks, never a count of people.
- A daily cap limits one profile. Ten profiles have ten caps.
keep reading
More on AI answer tracking cost
questions
AI answer tracking cost FAQ
Straight answers, including what Skald does not do.
Who pays for the AI answers?
You do, on your own accounts with OpenAI, Perplexity, Google and Anthropic. Each answer costs the model's tokens plus the web search the provider bills. The answers are asked with your keys, so they show on your provider bills, not on Skald's.
How does Skald estimate the cost?
Questions × tries × assistants with a key × each assistant's price an answer × runs a month (30 for daily, about 4.3 for weekly, 1 for monthly). The estimate shows on the profile's panel before you save.
What is the daily spending cap?
Each tracking profile has a cap in US dollars. Before each ask, the day's estimated spending plus that ask must stay under the cap. A run that would pass it stops and goes on the next day, and the page says how many days the run will take.
Is the estimate my bill?
It is a guide. The estimate uses rough default prices that your server's admin can match to your contract. Your real bill is on each provider's own usage page.
Does anything cost money before I set it up?
No. Nothing is asked, and nothing costs, until there is a key for at least one assistant and a tracking profile. An assistant without a key is left out, and the page says so.
Where do the API keys go?
In Settings › Integrations › AI answers, or in environment variables, the settings a server hands to Skald when it starts. Each key is write-only: it is never shown, printed, logged, or sent anywhere but to its own provider. How tracking profiles work.
$ tail -f access.log | skald
See what AI does with your site. Free for 30 days.
No card. We set up hosted Skald for you within one working day, or you self-host it free. Every line of your log, sorted and added up.