Brave Search in 2026: Every Change That Matters If You Collect Its Results

If you pull results out of Brave Search, 2026 has been a busy year. Brave's API lost its free plan in February. Its results pages got stricter about bots over the summer and autumn. And on 11 November, Brave announced that its result links are about to stop pointing at their destinations at all.

Each change was announced in a different place: a pricing page, a help article, a blog post, a GitHub thread. This guide puts them in one dated timeline and explains each in plain language first, then in technical detail. Then we cover what each change means for SEO tools, AI agents, researchers and developers, and what your options are.

One note on method. Every fact below comes from a primary source we read on 12 November 2026: Brave's own blog, help, pricing and documentation pages, Google's statement as reported by Search Engine Roundtable, and the public SearXNG issue tracker. Each one is linked where it is used. We did not send test traffic to Brave's results pages for this article. Where Brave hasn't said something, we say so rather than guess.

The 2026 timeline at a glance

Here is everything in order. A few late-2025 entries are included because they explain what came after. Each row links to the source it came from.

Brave Search changes that affect data collection, late 2024 to November 2026. Read from the linked primary sources on 12 November 2026.

The "Summer 2026" row is approximate on purpose. Search Engine Roundtable says the Google change began as a test in July 2026, and other reports date the first sighting to late June. Google's confirmation date is exact.

11 November: Brave's anti-scraping announcement

This is the change that made a guide like this necessary, so start here.

What happened, in plain language

Google made its own results harder to scrape. Scrapers that sell search results to chatbots and AI agents needed somewhere else to get results, and Brave runs one of the few independent web indexes. So they moved to Brave.

Brave says the result was "traffic spikes of several thousand requests per second", which it says used up its servers and its team "in roughly equal measure". To cope, its bot protection had to get stricter. Brave is open about who pays for that: "the bill for that falls on real users in the form of slower results and more CAPTCHAs."

Brave's answer is to change how result links work, so that collecting results in bulk becomes expensive while normal searching stays the same. The post is on Brave's blog, signed by the Brave Search Team and dated 11 November 2026.

Why Google's change landed on Brave

Over the summer, Google began routing clicks through google.com/goto. Hover over a Google result and you see a /goto address instead of the page itself, so the destination is no longer sitting in the page's markup. A Google spokesperson told Search Engine Roundtable on 26 August 2026: "We have a long history of deploying technical measures against evolving forms of abuse." Google didn't say who it was targeting.

Brave goes further than Google did about the motive. It argues that Google had tolerated scrapers for years, which quietly hurt rivals like Brave and Bing, and that scrapers now feed Google's AI competitors. That is Brave's opinion, not an established fact, but it explains why it took a firm line.

Brave's footnote on the alternative is also worth knowing. It describes Google's public API as one that "returns different results and is severely limited in usage". It then points everyone who needs results at scale to its own API, which it says offers the same results as Brave Search.

The new go/v1 result links, explained

Plain language first

Think of each result on a Brave page as a sealed envelope. The address of the website is inside. Each envelope has its own unique seal. When you click a result, your browser quietly asks Brave for the key to that one envelope, opens it, and takes you to the site.

To a person, nothing looks different: you click and the page opens. A program that reads the page's HTML, though, sees a stack of sealed envelopes and no addresses. Opening them all means asking Brave for a key every time, and Brave can count those requests for each results page.

The technical detail

• Link shape. Result links take the form search.brave.com/go/v1#…, with the encrypted destination after the #. Anything after a # stays in the browser, which is part of why the destination never passes back through Brave's servers.
• Per-link keys. Each link has its own key, and only Brave holds the keys.
• Blinded key requests. Your browser hides its key request behind a random value only it knows, so every request looks the same to Brave. Brave applies its secret, the browser removes the blinding, and it ends up with the one key it needs. It then decrypts the destination and navigates straight there.
• The cryptography. Brave names it as a partially oblivious pseudorandom function and links to RFC 9497. In Brave's words, it can tell that someone opened a link from a given results page and when, but not which link.
• Why it hurts bulk collection. Key requests can be counted per results page. One person opening two results makes two requests. A scraper that wants every destination on every page has to make one request per link, and Brave can see that pattern.

The side effects Brave lists

• Copied links aren't destinations. If you right-click and copy a result, you copy the redirect. Brave warns: "Redirect URLs might expire, so they should not be trusted as bookmarks or for long-term storage."
• Phishing risk. Brave admits a redirect "can become a vector for phishing attacks". Its safeguards are the ability to revoke a redirect, a time-to-live count, and an interstitial warning page.
• No JavaScript, no privacy. The scheme runs in the browser, so it can't work with JavaScript turned off. Those users get a conventional redirect that Brave's server can see.

What Brave hasn't said yet

Be careful with anything you read that is more specific than Brave's own post. As of 12 November 2026, Brave has not published:

• an exact date or region-by-region rollout plan (only "in the upcoming days");
• how long a redirect stays valid;
• whether the visible web address printed under each result title (the breadcrumb) changes too;
• any public specification for the key-request step.

We'll update this section when Brave publishes more.

The proof-of-work bot check

What it is

Brave Search has used CAPTCHAs for some time to stop bots scraping its results, while giving wrongly flagged people a way through. Its current design is a proof-of-work check. Rather than asking you to click traffic lights, your device solves a maths problem in the background, which Brave says was inspired by Bitcoin's proof of work.

Brave's help page gives the reasons. Picture CAPTCHAs fail too often and waste people's time. The new design needs "little to no user interaction" while still making scraping expensive for bots. Brave also says it relies on no third party and no private user information.

How it appears to people

The wording has changed over time. People have reported:

• "Verifying you are not a bot, quick check before you continue searching." This is the wording in a May 2026 report from a Pixel phone, where the check never finished.
• Earlier versions with a "Confirm you're not a robot" prompt and an "I'm not a robot" button, which appear in older Brave community threads.
• Occasional picture CAPTCHAs instead of the proof-of-work check. Some VPN users reported these on every query for about a day in November 2025.

Who gets it

Brave says only that people are flagged by its bot detection, and it lists no triggers. Its help page admits that people on popular VPNs are sometimes misclassified. Privacy Guides users report that the check depends on WebAssembly, so browsers or privacy modes that turn WebAssembly off (iOS Lockdown Mode, for example) can get stuck. After 11 November, expect the check to appear more often, because Brave has said its protection is stricter.

One point is non-negotiable here: this article won't tell you how to get past the check, and you shouldn't build anything that does. The check exists to make bulk collection expensive. If you keep hitting it on a project, that's the cue to switch to an API, covered below.

How to recognise each change

If you run a pipeline, a dashboard or a self-hosted search instance, this table maps each symptom to its cause. The most common mistake is recording a bot check as "no results". A check page can come back with an ordinary-looking status code, so judge what the page contains, not the status.

Symptoms and their causes. From Brave's pages, Brave's API documentation and public SearXNG issues, read 12 November 2026. Where a status code is unpublished, the cell says so.

If you hold saved Brave pages (from an archive, a research crawl or a monitoring job), here is a small guard that stops the first two cases from entering your dataset as if they were results. It only reads HTML you already have and sends no requests. We ran it on Python 3.14 on 12 November 2026 against the sample strings shown; it isn't a parser for Brave's results.

import re

# Wording Brave's bot check has shown to people (reported 2024-2026).
CHALLENGE_PHRASES = (
    "verifying you are not a bot",
    "confirm you're not a robot",
    "confirm you are not a robot",
)
# The result-link shape Brave announced on 11 Nov 2026.
GO_LINK = re.compile(r"https://search\.brave\.com/go/v1#[^\"'\s<>]+")


def brave_page_problem(html: str):
    """Return why a saved Brave page must not be recorded as data, or None."""
    text = html.lower().replace("’", "'")  # curly apostrophe -> straight
    if any(phrase in text for phrase in CHALLENGE_PHRASES):
        return "bot_check"            # not 'zero results': a gate
    if GO_LINK.search(html):
        return "encrypted_links"      # destinations are not in the HTML
    return None


samples = {
    "bot check": "Verifying you are not a bot",
    "older check": "<h1>Confirm you’re not a robot</h1>",
    "go links": '<a href="https://search.brave.com/go/v1#Zm9vYmFy">A result</a>',
    "neither": "<html><body>Nothing either check recognises</body></html>",
}
for name, html in samples.items():
    print(f"{name:11} -> {brave_page_problem(html)}")

Output:

bot check   -> bot_check
older check -> bot_check
go links    -> encrypted_links
neither     -> None

The habit behind this matters more than the code: a page that looks like "no results" should never be stored as "no results" until you know why it's empty. We wrote up the general version of that failure in why SERP scrapers break at 3 a.m.

The official Brave Search API in 2026

Brave's answer to "how do I get results at scale?" is its API, so here is what it looks like today. We read brave.com/search/api on 12 November 2026.

Brave Search API plans. Search, Answers and Enterprise from brave.com/search/api (read 12 Nov 2026). Spellcheck and Autocomplete from Brave's 12 Feb 2026 launch post.

Arithmetic is ours: $5 of credit at $5 per 1,000 requests is about 1,000 Search requests a month. Brave sells prepaid credits with no subscription, and a card is needed to activate a plan "to prevent abuse". The API page claims SOC 2 Type II.

Why you'll see three different "free" numbers

• $5 a month is the per-plan credit on each plan card. For web search, this is the number that counts.
• $10 a month is how Brave's 11 November post describes it: "$10/month in combined free credits; paid beyond that." Brave doesn't say which plans that combines; two plans at $5 each is our reading.
• Up to $20 is the banner on the API page. Brave's February post explains it: $5 on each of four plans "would be $20 across all plans".

The same 11 November post calls the API "free for personal and hobby use and paid for commercial use". That fits the credit model: a hobby project may never use up its $5. The February post also tied the credit to attributing the Brave Search API on your website or about page. Our guide to the free Brave Search API goes through the credit, the card requirement and the attribution question in detail.

Rate limits, in Brave's words

Per Brave's rate-limiting guide, limits are counted over a one-second sliding window, with a monthly quota reported over a 30-day window. Four response headers tell you where you stand: X-RateLimit-Limit, X-RateLimit-Policy, X-RateLimit-Remaining and X-RateLimit-Reset. Go over the limit and you get a 429 and the request fails. Brave says "Failed requests don't count against your quota", so refusals aren't billed.

What one request returns, and what it doesn't

The web search documentation sets these limits:

• Up to 20 results per request (count, default 20), and you may get fewer.
• An offset of at most 9, and you should only ask for the next page when query.more_results_available is true. Each page is a separate billed request.
• Each result has a title, URL and description, plus up to five extra snippets with extra_snippets=true.
• Local results come with IDs for a follow-up places lookup. Those IDs expire after about eight hours, so don't store them.
• Rich results (weather, stocks, sports) come back as a hint plus a callback key for a second call. Some of those data providers require attribution.

The web search documentation doesn't describe People Also Ask, ads or related-search blocks. Those are visible on Brave's results page, so if your work depends on them, check before you build. Brave also now recommends its LLM Context endpoint for agents and chatbots. It returns extracted, ranked page content rather than links, sized from about 1,000 to about 33,000 tokens, and each source carries a URL and timestamp.

The API page also describes Brave's index as "crawled, ranked, and served end to end".

What the API terms allow

Most comparison posts skip the terms, but they decide what you can build. The Brave Search API Terms of Service were last updated on 1 September 2026. The clauses that matter most for data work:

• Storage. You may not "store, cache, or create a database of Search Results, in whole or in part", except for transient storage needed to run your application. The API page's FAQ says storing results or training models is possible "with a custom plan".
• Resale. You may not "redistribute, resell, or sublicense the Search Results".
• AI. You may not use results to "create, evaluate, train, re-train, fine-tune, benchmark" AI models or services. The word benchmark matters if you run evaluation suites.
• Limits. You may not "circumvent or bypass rate limits or service limits through any method, including by creating multiple accounts".
• Attribution. The terms make it optional ("may"). If you add it, it must be conspicuous, such as "POWERED BY BRAVE" with Brave's logo. The free credit itself was tied to attribution in February.
• Ending. Brave can end the agreement "for any reason upon 10 days prior written notice".

We're not lawyers, and this is a summary, not advice. The practical point: if your project keeps an archive of results, for example a rank history or a research corpus, budget for the custom plan or another route, because the standard plans don't allow it.

What SearXNG ran into in October and November

SearXNG, the popular open-source metasearch engine, has two ways to query Brave. The brave engine reads the results page (web, news, images, videos and Goggles). The braveapi engine calls the official API and won't start without an API key. Both had public trouble this autumn, and the issue threads are a useful record of what Brave changed.

• Issue #6819, opened 5 October 2026. On version 2026.10.4, a single !brave test query ended in "Too many requests (suspended_time=180)". Contributors found two separate problems. Brave was refusing some requests outright. And even successful responses had changed shape: Brave's embedded result data now came wrapped in a JavaScript function (an IIFE), so the parser found nothing in a 200 response. The fix in #6836 merged on 19 October.
• Issue #6891, opened 28 October 2026. On image 2026.10.27, every braveapi search failed with HTTP 422 and the message "Input should be 'no-cache'". The HTTP library added Cache-Control: max-age=0 to each request, and the API strictly validates that header. Setting it to no-cache fixed it (PR #6910, closed 2 November).
• A standing warning in SearXNG's own docs: requesting too many pages can get you "flagged as a bot", with no results past about page 9.

Two lessons apply well beyond SearXNG. First, results-page integrations break silently: a 200 with nothing in it looks exactly like a quiet query. Second, the official API is strict about the requests it accepts, so if you're getting 422s, check your headers before you suspect your key. We tried running SearXNG as a free SERP API ourselves, and the honest verdict covers what happens to its engines from a single server.

Index, Goggles and AI answers

The anti-scraping news makes it easy to forget why so many people want Brave results in the first place. Briefly:

• An independent index. Brave describes its index as "one of only three independent, global-scale search indexes in the western world" (February 2026). It is built by its own crawler plus the opt-in Web Discovery Project. Brave no longer publishes a page count; its API page calls index size "a vanity stat". For data users, the point is simple: Brave's results are a genuinely different second opinion from Google's. We measured how different in how much search engine results overlap, and the engine-by-engine view is in Brave vs Google as a SERP data source.
• Goggles. Goggles are re-ranking rules that anyone can write and host on GitHub, Gist or GitLab. Brave still labels them beta. They're included in the API's Search plan, which makes them one of the few ways to get a custom-ranked independent index through an API.
• AI answers. The results page can show an AI summary ("Answer with AI"). Ask Brave (since 29 September 2025) adds longer answers, follow-ups and a chat mode; you trigger it with the Ask button or by ending a query with ??. Brave said in February that LLM Context powers "over 22 million answers per day" in Brave Search. If you track how AI answers describe your brand, that summary block is a separate thing to measure from the ranked links.
• News, images and video. Brave's February post puts them inside the API's Search plan at the same per-request price as web search. That's worth knowing when you compare per-request prices across vendors.

Three ways to get Brave results, compared

After 11 November there are three realistic routes. Here they are side by side, with each cell drawn from that route's published pages.

Ways to get Brave Search results, November 2026. Brave figures from brave.com/search/api and Brave's documentation; Serpent figures from our pricing page and Brave Search API page. Read 12 November 2026.

Other SERP API vendors: when we checked the seven largest wholesale vendors on 16 September 2026, six published no Brave product and one sold a Brave AI Mode product rather than web search. The full survey, with sources, is in who resells the Brave Search API.

A few honest notes on reading the table. Brave's own API is the source of record and costs nothing if you stay inside the monthly credit. No per-call price beats free, and if a thousand requests a month covers your job, start there. LLM Context and Goggles are Brave-only features you won't get elsewhere.

The DIY column is in the table because people ask about it, not because we recommend it. Brave has now said publicly, in its own words, that it is making bulk collection from its results page expensive. Building a business process on top of that is a bet against Brave's engineering team. For the wider version of this trade-off, see web scraping vs a SERP API.

What this means for you

If you run an SEO or rank-tracking tool

Rank tracking needs two things per row: a position and a URL. Once the go/v1 links reach your markets, the results page's HTML no longer carries the URL. Check where your Brave numbers come from. If your vendor gives you JSON with a url field, you have what you need. If you read the page yourself, expect the URL column to go empty or fill up with go/v1 strings. Either one would quietly corrupt a rank history, so add a check now. Our post on finding domains in Brave results shows why keeping the query, URL and host together matters.

If you build AI agents or RAG pipelines

Brave is positioning its API squarely at you: LLM Context, cited Answers and an OpenAI-compatible chat endpoint. Two cautions. First, the terms forbid using results to train, fine-tune or benchmark AI models, so an evaluation harness built on Brave results needs a conversation with Brave first. Second, if your agent "browses" Brave's results page as text, it will see go/v1 links rather than destinations, and it may hit the bot check. Give it an API instead. Our comparison of web search APIs for agents and RAG sorts the options by the output your agent actually needs.

If you're a researcher, journalist or analyst

The change most likely to bite you is citation links. When you copy a result from Brave, you'll now copy a redirect that may expire. Open the page and copy the address you land on. If you're building a dataset, mind the storage clause: the API's standard plans don't allow keeping a database of results. For cross-country or longitudinal studies, our guide to comparing Brave results across countries covers sampling design.

If you self-host SearXNG or another metasearch tool

Update to a build that includes the October and November fixes, and expect the brave engine to be suspended more often from server addresses as Brave tightens up. The sanctioned option inside SearXNG is the braveapi engine with your own key, which the monthly credit may cover for personal use.

If you already use the official API

Nothing in the 11 November post changes the API. Brave is pointing people towards it. Watch the X-RateLimit-* headers, treat a 422 as a request-format problem, and re-check the pricing page before you budget, because the plan line-up has changed twice since August 2025.

If you just search on Brave

Clicking results works as before, and Brave designed the new links so it can't see which result you opened. You may see the bot check more often for a while, especially on a VPN. If it never finishes, check that JavaScript and WebAssembly are allowed for search.brave.com.

Brave results through Serpent

If you'd rather call one API than maintain anything, Serpent returns Brave results as structured JSON through the same endpoint and API key you'd use for Google, Bing, Yahoo and DuckDuckGo. Set engine=brave on /api/search and you get:

• Quick Search for the organic list (position, title, URL, snippet, display URL), one call and one charge, built for rank tracking.
• Deep Search for the full results page: People Also Ask with answers and source links, related searches, ads, rich snippets, knowledge panel and local pack where Brave shows them, and, on paid plans, Brave's AI digest as aiOverview when you add include_aio=true.
• News, image and video results from Brave on their own endpoints.
• A fixed key set. A block Brave didn't show comes back as null or [], never as a missing key, so one parser covers every engine. We wrote up that pattern in normalising SERP JSON across providers.
• Country and language targeting across 112 country codes, with freshness and safe-search filters.

Pricing is published: web search runs from $0.60 per 1,000 calls on Default down to $0.03 per 1,000 on Enterprise Plus, and every engine is billed at the same rates, so switching engine= never changes your bill. New accounts get 10 free calls. Everything is on the Brave Search API page and the documentation, and you can run a Brave query in the playground without signing up. To query several engines at once, the multi-engine aggregator walkthrough shows the pattern.

FAQ

Why does Brave Search keep asking me to verify I am not a bot?

Brave runs a bot check on its results pages, and on 11 November 2026 it said that a flood of scraper traffic had forced it to make that protection stricter, with slower results and more CAPTCHAs for real people as the side effect. The usual check is a proof-of-work challenge: your device solves a small maths problem, and the screen reads along the lines of "Verifying you are not a bot, quick check before you continue searching." It needs JavaScript and WebAssembly. Brave's help page says people on popular VPNs are sometimes misclassified as bots.

What are search.brave.com/go/v1 links?

They are Brave's new result links, announced on 11 November 2026 for rollout "in the upcoming days". Instead of pointing straight at the destination, each result points at search.brave.com/go/v1# followed by the destination in encrypted form. Each link has its own key, only Brave holds the keys, and your browser fetches the one key it needs when you click, in a way that stops Brave seeing which link you chose. For a person nothing changes. For a program reading the page's HTML, the destination URL is no longer there to read.

Can I still bookmark or share links copied from Brave Search?

Not reliably once the go/v1 links reach you. A copied result link is the redirect, not the destination, and Brave says redirect URLs might expire and should not be trusted as bookmarks or for long-term storage. Open the result and copy the address of the page you land on instead. That matters most for researchers and writers who keep citation lists.

Is the Brave Search API still free in 2026?

There is no free plan any more, but there is free credit. Since 12 February 2026 each Brave Search API plan includes $5 of credit a month. On the Search plan, at $5 per 1,000 requests, that covers about 1,000 requests. Brave's launch post tied the credit to attributing the Brave Search API on your site, and a card is needed to activate a plan. Brave's 11 November post describes $10 a month of combined credit, and the API page banner says up to $20, because the credit is counted across several plans.

Does the Brave Search API return the same results as search.brave.com?

Brave says it does: its 11 November 2026 post states that the Brave Search API offers the same results as Brave Search. The shape differs. The web endpoint returns up to 20 results per request with an offset of at most 9, and the documentation does not describe People Also Ask, ads or related-search blocks. The results page shows things the API documentation does not mention, and the API offers things the page does not, such as the LLM Context endpoint.

Is scraping Brave Search allowed?

Brave's 11 November post treats bulk scraping of its results pages as the abuse its protection is built to stop, and points anyone who needs results at scale to its API. The API's own terms, updated 1 September 2026, bar getting around rate limits by any method, including multiple accounts. We are not lawyers. If your project depends on Brave data, read Brave's terms and take advice, then use an API: Brave's own, or a SERP API that publishes its own terms.

For a shopping-side view of the same engine, see what Brave product results actually contain.