DeepSeek answers API

What the DeepSeek API returns

DeepSeek's model doesn't search the web by itself. We run DeepSeek's own search for the question, give the model the results, and it answers from them with numbered citations. Each answer comes back read the same way as the other engines: brands in order, every citation typed, and the raw response kept as proof.

One question, three different answers

We asked “best invoicing app for freelancers” on each engine from the US on the same day. Each card is what the API returned: how the answer opens, the brands it named in order, and the first sites it cited.

Where each answer ranks each brand (1 = named first; – = not named)

Brand	DeepSeek	ChatGPT	Google AI Overview
Wave	1	3	1
Zoho Invoice	2	1	2
Bonsai	3	4	– not named
FreshBooks	4	2	3
Stripe Invoicing	5	– not named	– not named
Square Invoices	6	– not named	– not named
PayPal Business	7	– not named	– not named
Invoice Ninja	8	5	4
Invoice Simple	– not named	– not named	5

All three name Wave, Zoho Invoice and FreshBooks, but DeepSeek puts Wave first while ChatGPT leads with Zoho Invoice. DeepSeek also names payment tools the others leave out: Stripe Invoicing, Square Invoices and PayPal Business. It cites none of the same sites as ChatGPT or the AI Overview.

Calling the DeepSeek API yourself vs aivisibilityskill

DeepSeek's chat API answers from what the model already knows, with no live sources. To see what DeepSeek says about a market today, the answer has to be grounded in a fresh search, and then read for brands. That's what this API does, in the same schema as ChatGPT, Gemini, Google AI and the rest.

Calling DeepSeek yourself vs through aivisibilityskill

	DeepSeek chat API	aivisibilityskill
Live web sources	Not by itself	DeepSeek's own search, cited in the answer
Brands named	You parse them yourself	Listed with rank, recommendation and sentiment
Source types	No	Every citation typed
Billing	Tokens	One price per answer; failed answers are free
ChatGPT, Gemini, Google AI and others	No	Yes, in the same call

To be clear about what this measures: it's DeepSeek's model answering from DeepSeek's search for a fresh question. The DeepSeek app runs its own search and settings, so its answers can differ.

Track brand mentions in DeepSeek

DeepSeek is the cheapest of the model engines to ask, so it fits daily tracking next to ChatGPT and Google AI Overviews. Put a brand, its competitors and its buyers' questions in a project with a schedule, then read how often DeepSeek names the brand (with error bars), competitor share, cited sites and history. A webhook tells your server when each run is done.

Call the DeepSeek API from TypeScript or Python

The SDKs wrap the same request, retry safely without charging twice, and wait for the answers for you. Free sandbox keys (aiv_test_…) return realistic sample answers, so you can build the whole feature before spending a credit.

TypeScript (npm install @aivisibilityskill/sdk)

import { Aiv } from "@aivisibilityskill/sdk";

const aiv = new Aiv({ apiKey: process.env.AIV_KEY! });

const job = await aiv.checks.create({
  questions: ["best invoicing app for freelancers"],
  engines: ["deepseek"],
  locale: "en-US",
});
const done = await aiv.jobs.wait(job.id);

for (const answer of done.answers ?? []) {
  console.log(answer.brands?.map((b) => `${b.rank}. ${b.name}`));
  console.log(answer.citations?.map((c) => c.domain));
}

Python (pip install aivisibilityskill)

import os
from aivisibilityskill import Aiv

aiv = Aiv(api_key=os.environ["AIV_KEY"])

job = aiv.checks.create(
    ["best invoicing app for freelancers"],
    engines=["deepseek"],
    locale="en-US",
)
done = aiv.jobs.wait(job["id"])

for answer in done["answers"]:
    print([(b["rank"], b["name"]) for b in answer["brands"]])
    print([c["domain"] for c in answer["citations"]])

For thousands of questions, send them as one batch of up to 20,000 answers and get one webhook when it's done. Full guides: SDKs, API reference, OpenAPI 3.1 spec.

Prefer a ready-made tool? llm-brand-tracker on GitHub tracks your brand against competitors on all 8 engines every week, free on GitHub Actions. Free and MIT licensed.

DeepSeek API pricing

You pay per answer, in credits. DeepSeek is asked live, one call per answer, so there's one speed. A dollar buys 1,000 credits on a pack; monthly and volume plans include more. An answer reused from a recent identical one (max_age_hours) costs half, and an answer that fails costs nothing.

Every account gets the same API and every engine, starting with 500 free credits and no card. Volume plans start at $500 a month for 750,000 credits, with invoicing and a 99.5% uptime commitment: see volume plans.

Countries and languages

DeepSeek is asked through its API, so it isn't answering from a location. For any locale other than en-US, it's told which country the buyer is in, for any of 250 countries and territories, and it answers in the question's language (de-DE, pt-BR, ja-JP…). The app and search engines answer from the country itself: ChatGPT in 192, Gemini, Google AI Overviews and AI Mode in 240. GET /v1/locales lists every pair for free.