Changelog

New features, integrations, and improvements to AI Data Stream

November 2026

Integration

Gemini 4 Hydrogen, Carbon and Argon now available

Google’s three Gemini 4 models are now in the model picker. Gemini 4 Hydrogen ($1.50 / $7.50 per million input / output tokens) is the fast, efficient one and is also on Google’s free tier. Gemini 4 Carbon is Google’s strongest reasoning model, and Gemini 4 Argon is the faster option at the same price: both are $2 / $10 through 30 November 2026, then $4 / $20, and need a paid Google API key. Gemini 3.8 Flash stays the default and Gemini 3.1 Pro stays as the previous generation. Gemini 2.5 Pro has been retired because Google will answer its requests with Gemini 3.8 Flash from 1 December 2026: if it was your default, Gemini 4 Hydrogen is now used automatically.

Read more
Integration

Claude Fable 5.2 now available

Claude Fable 5.2 is now in the model picker, and it costs less than the model it replaces: $8 / $40 per million tokens, down from $10 / $50 for Fable 5.1, with cache reads at $0.20 per million. Anthropic reports large gains in reading charts and up to 30% faster output than Fable 5.1. Fable 5.1 and Fable 5 are retired, and anyone using either moves to Fable 5.2 automatically. Sonnet 5.5 is still the default. As before, Fable 5.2 relies on 30-day data retention on most Anthropic accounts; Anthropic says eligible enterprise customers can use it with zero data retention.

Read more

October 2026

Improvement

Clearer labels on brand mentions

Stored brand mentions are now read automatically in the background to see what kind of page each one is and how it treats your brand. A mention keeps a rougher label until it has been read.

  • New labels: Critical and favourable mean the page says something about the company itself, and just named means it only names you, for example as a seller
  • Sales listings hidden: Pages recognised as sales listings are hidden by default. Turn on Show mentions hidden by noise filters to see them
  • Filter from the tile: Click a count on the Favourable / critical tile to filter the list, or use About the brand
  • Automatic: The labels are automatic, so read the text before acting on one
Read more
New Feature

Brand mentions page

Brand mentions now have their own page in the property sidebar, so you can browse what’s stored without asking the AI. Tick Brand mentions on your DataForSEO connection and add a brand term or competitor to fill it.

  • Filter the list: By term, date range, page type, text, site and sentiment, with republished copies grouped under one mention
  • Top sites: See who mentions you most and click a site to show only its mentions
  • Free to browse: The page reads what’s already stored and never calls DataForSEO
  • Noise filters on the page: Property admins can edit Exclude mentions containing and Exclude domains, or exclude a site with one click. Show mentions hidden by noise filters shows what was left out and why
  • Check for new mentions: Property admins can fetch what DataForSEO found since the last check for the selected terms, at about $0.03 to $0.06 per call and at most 3 calls per term. Terms checked in the last hour are skipped
New Feature

Stored backlinks and a Backlinks page

Backlinks are now stored for the domains you track on your DataForSEO connection, so asking again, filtering and sorting cost nothing. Tick Backlinks under Tools the AI can use and add your domains.

  • Tracked domains: Up to 2 of your own, filled in from Search Console with one click if connected, and up to 5 competitors. Any other domain or a single page URL is still a live lookup
  • Refreshed when used: A domain is refreshed when the AI uses it and its links are more than 7 days old, with a full refresh every 30 days. Nothing is spent on a domain nobody uses
  • Weekly refresh, opt in: Turn on Refresh backlinks weekly to keep your own domains current. For a site with about 5,000 links that costs roughly $0.08 a week plus $0.40 for the monthly full refresh
  • New and lost links: Stored links stay after they disappear from the web, so the AI can say which links you lost and when, and show new and lost links per month
  • Backlinks page: A new sidebar page lists the stored links with status, rank, spam score and date found, plus your top referring domains and anchors. Browsing costs nothing, and property admins can press Refresh now
  • Big sites: Sites with more than 20,000 links are stored as a sample of the strongest and newest links, and anchor lookups for them stay live
Read more
New Feature

Brand mentions with DataForSEO

The DataForSEO connection can now track where your brand is mentioned across the web. Tick Brand mentions under Tools the AI can use, add your brand terms, and ask your AI what was said about you this week, who said it, and how coverage compares with last quarter.

  • Two new tools: The mentions themselves, from news, blogs and forums by default, with the source, a snippet, the date DataForSEO found the page and a rough sentiment, plus mention trends per day, week or month
  • Republished copies counted once: A press release picked up by a dozen sites is one story with a count, not twelve mentions
  • Competitors: Add up to 5 so the AI can compare share of voice without being told who to compare against
  • Paid for once: Mentions are stored, so asking about the same period again costs nothing and later questions only pay for pages found since. Each lookup makes at most 3 metered calls, and the AI says what a lookup cost
  • Daily sync, opt in: Turn on Sync brand mentions daily and new mentions of your brand terms are fetched once a day
  • Noise filters the AI can tune: Every brand has its own noise. The AI points out sites and phrases that keep showing up without being real coverage and, if you agree, adds them to Exclude mentions containing or Exclude domains. Filters apply to stored mentions straight away, at no cost (a brand stored as a sample is fetched again)
  • Stored mentions in the settings: The dates stored for each term, the last sync and what it has cost, with Remove stored data
  • Brand Mentions Digest skill: A round-up of the latest coverage, sorted into press, blogs, forums, republished copies and noise. On paid plans, schedule it for a weekly email
  • All plans, bring your own key: Calls are billed to your own DataForSEO account at around $0.03 each, or about $0.06 for a full page of 1,000 results
Read more
New Feature

Web search for DeepSeek providers

DeepSeek providers can now let the AI search the web and read web pages, for questions your connected data can’t answer, such as algorithm updates, industry news or a competitor’s page. The AI cites the pages it uses. It’s off by default: turn on Enable web search and page reading under Edit provider in Team Settings → AI Providers. DeepSeek bills your own account $0.75 per 1,000 successful searches and $0.75 per 1,000 pages read.

Integration

OpenAI GPT-6.1 Sol and GPT-6.1 Terra now available

GPT-6.1 Sol and GPT-6.1 Terra are now in the model picker. GPT-6.1 Sol ($2 / $10 per million input / output tokens) is the same price as GPT-6 Sol and becomes the default for newly added OpenAI providers. GPT-6 Sol stays as the previous generation. GPT-6.1 Terra ($1 / $5) costs half as much as Sol and sits between Sol and GPT-6 Luna. GPT-6 Astra remains the premium option. Two older models have been retired: if GPT-5.6 Sol was your default, it now uses GPT-6.1 Sol at half the price. GPT-5 Mini, which OpenAI shuts down on 11 December 2026, moves to GPT-6 Luna, which costs less. Both new models offer every reasoning effort level from Low to Max.

Read more
Integration

Sitemap connection — know which pages appeared and disappeared

Sitemap is now a connected data source on Pro and Business plans. Point it at your sitemap.xml and AI Data Stream keeps a history of every URL it lists, for any site that publishes a sitemap: static builds, Shopify, headless CMSs, anywhere WordPress can’t reach. No credentials needed.

  • Change history: Which URLs appeared, which disappeared and when, with each removal reported as an honest window between two fetches rather than a guessed day
  • Traffic left behind: Take the removed URLs to Search Console or GA4 and see how many clicks walked out with the deleted pages. The new Sitemap Content Changes skill does exactly that
  • Date a disappearance: Ask when one page left the sitemap, or run a list of 404s through it to sort real deletions from junk URLs
  • Background refresh, opt in: By default the sitemap is only fetched when someone asks. Turn on Refresh in the background and it is checked every refresh interval, so removal windows shrink to a day. Checks are conditional and only download what changed
  • lastmod quality: The connection flags sitemaps with no lastmod, or with the same date stamped on every URL, since search engines ignore the field once it proves unreliable
  • Dashboard widget: URL counts, additions and removals over 7 and 30 days, recent changes and the largest sections
  • Sitemap page: A new page in the property sidebar to browse everything the AI sees: sections, changes over any date range, and a URL search that includes removed pages and when they went
  • Safe by design: Plan caps of 10,000 URLs on Pro and 50,000 on Business, a guard that refuses to mark thousands of pages removed when a fetch comes back broken, and read-only access to a public file
Read more
Improvement

Deeper Search Console, Core Web Vitals and PageSpeed tools

The Google tools your AI already had now use much more of what each API offers. No new connections to set up; existing ones pick the changes up automatically.

  • Discover, Google News, image and video search: Search Console performance can be asked for by search type, reported separately from web search. Content sites can finally see Discover traffic in chat
  • Filters and paging: Narrow any Search Console question to a page prefix or regex, a country, a device or a search appearance, and page through results beyond the first thousand rows
  • Fresh data: Ask about the last few days and the AI can include Google’s provisional data instead of stopping three days back
  • URL inspection, in full: Google’s chosen canonical versus yours, robots and fetch state, which sitemaps and pages Google found the URL in, rich results with their issues, and up to ten URLs in one call
  • Sitemap status: Which sitemaps you submitted, when Google last downloaded them, and their error and warning counts
  • Core Web Vitals, per page and per band: Query a single URL instead of the whole site, see what share of visits are good, needs improvement or poor rather than one percentile, and go back up to 40 weeks
  • Why LCP is slow: A new snapshot breaks LCP into time to first byte, resource load delay, load duration and render delay, says whether the largest element is an image, and shows how visitors arrive (fresh navigation, back/forward cache, reload, prerender)
  • PageSpeed diagnostics: The audit now names the LCP element and its phases, layout-shifting elements, render-blocking requests and third-party scripts by blocking time, reports real-user data for the page and the whole site side by side, and lists the failing SEO, accessibility and best-practice checks with example elements when you ask for those categories
New Feature

AI search visibility with DataForSEO

The DataForSEO connection now covers AI search visibility as well as backlinks. Ask your AI which Google AI Overview and ChatGPT answers cite your site, which pages earn those citations, and how competitors compare, from DataForSEO’s corpus of real AI answers.

  • Five new tools: Visibility summary with monthly history, the actual cited answers, the pages that earn citations, a side-by-side competitor comparison, and the domains cited most for a topic
  • AI Search Visibility skill: One report that gives you an honest baseline before anyone sells you GEO or AEO services, and a way to check their claims afterwards
  • Coverage: Google AI Overviews in around 94 countries and ChatGPT (United States, English); data since August 2025
  • All plans, bring your own key: Calls are billed to your own DataForSEO account at roughly $0.10 each; the existing row ceiling and allowed-domains guards apply
  • Plan feature renamed: “DataForSEO Backlinks” on the pricing page is now simply “DataForSEO”
Read more
Improvement

Connect Your Own AI on Every Plan

Connecting your own AI assistant (claude.ai, Claude Desktop, Claude Code and other MCP clients) is no longer limited to Pro and Business. Every plan now includes it, and it is switched on by default.

  • Free plan: 300 tool calls a month per team, reset on the 1st (UTC); listing properties, skills and the docs doesn’t count
  • Pro and Business: Unlimited tool calls
  • On by default: New teams start with External AI Access on, and team owners can still switch it off in Team Settings
  • See your usage: Free team owners see this month’s calls under Team Settings → External AI Access
New Feature

To-do List

Every property now has a shared to-do list for the follow-ups your analysis turns up. Add tasks yourself, or let the AI note them down in chat and tick them off when they’re done.

  • AI adds follow-ups: When a conversation turns up something to act on, the AI adds it to the list and says so in its reply
  • Never silent: An Added to to-do list chip in the chat header links straight to the list
  • Minimal list: Add, tick off, star and drag tasks, with optional due dates and notes, and a link back to the chat a task came from
  • Team-shared: One list per property; changes need Write permission, and incognito chats can read but not change it
  • Works with your own AI: Available through your personal connector too
  • Free plan: 25 open tasks per property; up to 500 on paid plans (completed tasks don’t count)
Read more
Improvement

GPT-4o and GPT-4o Mini retired

OpenAI is shutting down GPT-4o and GPT-4o Mini on 14 January 2027, so both have been removed from the model picker. Anyone who had GPT-4o selected is migrated to GPT-6 Sol automatically, and GPT-4o Mini moves to GPT-6 Luna, its closest match for speed and price. GPT-6 Luna ($0.10 / $0.50 per million tokens) is now the fastest OpenAI option, and it also takes over the quick background tasks that used GPT-4o Mini, such as naming conversations and summarising annotations. GPT-6 Sol ($2 / $10) remains the recommended OpenAI model.

Improvement

AI crawler counts now exclude failed fetches

Vulnerability scanners routinely borrow AI crawler names while probing for .env, key and config files, and on some sites those probes outnumber real crawler traffic several times over. The Cloudflare connection now separates the two: the dashboard widget and the AI crawler tools count successful fetches (2xx/3xx) as requests, and report 4xx/5xx responses as a separate failed total with a note on what they usually are. Bandwidth counts successful fetches only. Ask the chat for AI crawler paths with status 400 and above to see exactly what the scanners asked for.

Read more
Integration

DeepSeek Flash with thinking and effort levels

DeepSeek in AI Data Stream now runs on DeepSeek Flash (V4.1-Flash), DeepSeek’s current model, with thinking on and a Low / High / Max effort picker in chat, the same control Claude and GPT users already have. The old “Chat” and “Reasoner” entries are retired: DeepSeek dropped those names with V4, and anyone still on them is migrated to Flash automatically. Pricing is DeepSeek’s published rate, charged to your own key: $0.15 per million input tokens and $0.60 per million output tokens off-peak, double that at peak hours.

Read more
Improvement

Gemini 3.7 Flash retired

Google has deprecated Gemini 3.7 Flash and now routes every request for it to Gemini 3.8 Flash, so the 3.7 Flash entry has been removed from the model picker. Anyone who still had 3.7 Flash selected is migrated to 3.8 Flash automatically. Pricing is unchanged: both models share the same introductory rate ($0.75 / $3.75 per million tokens through 31 December 2026, then $1.50 / $7.50). Gemini 3.8 Flash remains the recommended Google model.

Integration

Claude Haiku 5.5 now available

Anthropic’s new Claude Haiku 5.5 is now in the model picker as the fastest Claude option. It costs $0.10/$0.50 per million tokens for prompts up to 100,000 tokens, around 90% less than Haiku 4.5, and it is the first Haiku with adjustable effort levels and a 1M-token context window. Background tasks on Anthropic providers (conversation titles, annotation and knowledge-base summaries, AI prompt generation) now run on it automatically. Haiku 4.5 stays available as the previous generation.

Read more
Improvement

Starter annotation categories and permanent deletion

Every team now has five starter annotation categories - Marketing Campaign, Site Change, Content Update, Product Launch and Bug Fix - so anyone with write access can add annotations straight away instead of waiting for the team owner to create a category. Teams that already had categories keep exactly what they had. Deleting an annotation still archives it, and you can now remove archived annotations for good with Delete permanently in the Archived view. The Team Settings tabs and the account settings menu are now translated into French and German.

New Feature

The AI now answers questions about AI Data Stream itself

Ask the AI how to do something in AI Data Stream - connect Search Console, add an AI provider key, invite a teammate, schedule a report - and it now reads the documentation before answering. Answers use the exact button and menu names you see, in your language, and link to the relevant section of the docs. If you ask for data from a platform that isn’t connected to your property yet, such as Meta Ads, the AI tells you which plans include it and how to connect it instead of just saying the data is unavailable. The same documentation is available to your own AI client through the AI Data Stream connector. We also checked every documentation page against the app and corrected outdated steps and labels in English, French and German.

September 2026

Integration

Claude Sonnet 5.5 now available

Anthropic’s new Claude Sonnet 5.5 is now the default Anthropic model in AI Data Stream. It costs the same as Sonnet 5 at $2/$10 per million tokens but generates 30%+ faster and lands within two points of Opus 5.5 on real professional work. Sonnet 5 stays available as the previous generation so you can compare. Sonnet 4.6 has been retired; if it was your default, you move to Sonnet 5.5 automatically.

Read more
Integration

OpenAI GPT-6 Sol and GPT-6 Luna now available

GPT-6 Sol ($2 / $10 per million input / output tokens) and GPT-6 Luna ($0.10 / $0.50) are now in the model picker. They replace GPT-5.6 Terra and GPT-5.6 Luna, which have been retired: if either was your default, it now uses GPT-6 Sol or GPT-6 Luna automatically, at a lower price. GPT-5.6 Sol stays as the previous generation, GPT-6 Astra remains the premium option, and GPT-6 Sol is now the default for newly added OpenAI providers. Both new models offer every reasoning effort level from Low to Max.

Read more
Integration

Claude Opus 5.5 now available

Anthropic’s new Claude Opus 5.5 is now the most capable Anthropic option at standard pricing, one tier below the premium Fable 5.1 — 20% cheaper than Opus 5 at $4/$20 per million tokens, with cache reads down 60%, while performing close to Fable 5.1. Opus 5 stays available as the previous generation so you can compare. Opus 4.8 has been retired; if it was your default, you move to Opus 5.5 automatically.

Read more
Integration

OpenAI GPT-6 Astra now available

GPT-6 Astra, OpenAI’s new flagship, is now in the model picker as a premium option at $10 / $50 per million input / output tokens - 2.5 times GPT-5.6 Sol at its promotional $4 / $20, with cached input at $1. Nothing has been retired: Sol, Terra and Luna all stay, and your default model does not change. Astra and the GPT-5.6 family now offer extra-high and max reasoning effort levels, the chat effort picker starts from the team default you saved for each provider, and our OpenAI integration has moved to OpenAI’s Responses API so reasoning effort applies fully while the model is calling your data tools.

Read more
New Feature

Jev by TypeSafe AI joins the app — a System One model for AI traffic classification (beta)

AI Data Stream now runs a second kind of model alongside the frontier LLMs on your own keys: Jev, the first System One model from TypeSafe AI. Jev never generates text. You send it a state and typed questions, and it returns a choice from your options or a yes/no probability, with a calibrated confidence, in about a second. It runs on our key, not your BYOK LLM key, and the blog post explains what a System One model is, how it differs from an LLM, and how we validated it.

Its first job is an experimental, switch-it-on-yourself beta for Pro and Business teams. GA4’s new “AI Assistant” channel only catches visits that arrive with an ai-assistant medium. Perplexity, Copilot and untagged ChatGPT links land in Unassigned instead, next to your own subdomains, link scanners and referrer spam. When the beta is on, every traffic-source lookup gives those rows a second opinion from Jev.

  • A real channel for Referral and Unassigned rows: AI assistant, self-referral, link scanner or tool, referrer spam, industry referral, search, social, email, offline or campaign tag — with a confidence score on every row
  • An is_ai_assistant flag your AI can total, so “how many sessions came from AI” finally includes the ones GA4 missed
  • GA4 stays the source of truth: below a confidence threshold the GA4 grouping is kept and the model’s guess is only noted; if the model is unavailable you get plain GA4 results, never an error
  • Judged once, cached 30 days per source, so the extra second only applies to the first lookup of each source
  • On/off in Team Settings → Beta Features (team owner, paid plans). Off by default. With it off, your AI mentions it once when your data has rows it could help with, with a link to the toggle
Read more