Every built-in tool in Fractl Agents.

AI models increasingly read your pages before any human does — summarizing you, answering questions about you, and forming an impression. This tool makes that impression measurable. It shows the page to a panel of leading frontier models and scores how well they understand it: who/what they think you are, whether your specific claims survive intact, what message survives compression, how the read matches your intended positioning, and what the models already 'remember' about the brand versus what the page actually says. Every score is self-teaching and comes with the exact misreads and a ranked, fix-each list.

Ask the AI engines (ChatGPT, Gemini, Claude, Perplexity, Google's AI) the same question repeatedly and capture what happens under the hood: the query map shows the hidden searches each engine ran (what to rank for, which list pages to be on, which brands get pre-selected by name), and the citation analysis compares the pages the engines cited against the pages they saw and skipped, with statistical backing and a concrete pitch-target list.

Measure brand mentions and citations in AI answers. Compare a saved AI visibility run with current search results for a fixed query set, retaining each source and observation date.

Build or audit content against search intent and source evidence. Reuse saved research to compare revisions, map entities to internal pages, or export claims and their support.

This is for deciding which pages should be updated, expanded, consolidated, internally linked, validated with first-party data, pruned, redirected, or handed into deeper content planning.

This tool starts upstream of content briefs. It helps the team decide what to write about by building a topic portfolio from seeds, competitors, category context, and opportunity scoring.

This is the tool for 'why is our traffic down' and 'what is actually holding this site back' — questions that need an answer with evidence behind it, not a checklist. It runs around 142 typed checks over the site's own search and analytics data, then makes a judge prove each finding directly from that data. A finding that cannot be evidenced is set aside and reported as such — never narrated as if it were true. It then works out the mechanism behind each finding, and chases whatever the first-party data cannot answer through live web search, the SERP and backlink data. It costs around $175 and takes several hours, so it always confirms the spend and the analysis window first.

This tool studies which pages in and around a client's space actually earn links from other websites, how many people search each month for tool-style answers, and which comparable tools are real working interactives versus articles dressed up like one. From that evidence it designs a ranked shortlist of interactive tools the client could publish - each concept differentiated from what exists, with a data hook and a complete build spec ready for the coding workbench.

This is a page-level technical check for performance, accessibility, best practices, and SEO signals.

Every client wants to be on the 'X Top Digital PR Agencies' or 'Best [category] Companies' lists — those editorial roundups are prime, durable backlinks. This tool finds them at scale: it works out the categories a company genuinely belongs on, discovers candidate roundup/listicle pages two ways (search-query probes like 'best digital PR agencies' and the pages that already link to close competitors), then reads and grades each page for whether the brand could realistically be featured — relevance, inclusion-fit, editorial quality, page type — and scores it by link value (domain + page authority, page traffic) and ease of acquisition. It resolves the author/editor contact for the best targets and returns a ranked outreach tracker: prioritized targets, worth-a-look, unlinked-mention easy wins, and lists you're already on. Marginal finds are kept in a labeled long-tail section, so nothing useful is dropped.

This tool looks at one domain at a time and shows which pages already earn organic visibility. It is useful for editorial planning and competitor research, but it is not a full multi-domain gap diff by itself.

A whole-site intelligence tool for SEO and content strategists. It crawls the site, reads every page with an AI extract, audits ~20 technical/SEO/content/link areas, maps the site's knowledge and distinctiveness, and (optionally) checks whether AI engines cite the site — then renders an interactive workspace dashboard. It is also the platform's general crawler: collect mode delivers raw page-content and link-graph CSVs from a domain or URL list, and watch mode monitors a site for changes over time.

This is an ideation and market-intelligence tool for content and PR teams. It validates trends across search, social, news, video, podcasts, and category-specific datasets, then turns them into concrete campaign angles.

Describe an audience as people, or use the active client's, and get the podcasts, YouTube channels, websites, press outlets, social accounts and subreddits that audience over-indexes on, the AI prompts they ask, and their demographics, each ranked by affinity. Saved per client, so a second look is free.

This is campaign measurement, not brand monitoring. Given the URL of one published study or data asset, it traces the coverage outward: the articles that wrote about it, the wire copies that republished those articles, the aggregators and translations that carried it further, and the later stories that cite the coverage rather than the study. It reads each page rather than trusting the search snippet, so every placement it counts comes with the sentence that proves it.

This is the influencer version of journalist discovery. Give it a topic, a client website, or a document and it finds the people who move opinion in that space, ranks them by how well they actually fit, merges each person across their channels, finds contact info, and builds a ready-to-pitch target list you can turn into a campaign.

This is a cross-channel headline and hook generator for PR, SEO/blog, Reddit, YouTube, X, and email-style formats.

When you want a deep dossier on one specific influencer/creator to partner with them well — who they are across all their channels, what they actually cover, their audience, and how to pitch/partner. Give a discovered influencer (identity_key) or a name + a handle/URL for a cold target.

A local/geographic ideation tool for content and digital-PR teams. From a brand's site URL it finds ownable 'best/worst metros/states for X' data studies and local content angles, proves every number against its public source, and generates a tailored pitch line for every market.

This is the media-research search path, not a general web search. It is designed to find articles, authors, outlets, beats, and recency-relevant coverage.

This is for researching one journalist deeply enough to pitch them intelligently. It is not just contact lookup; it builds the context a PR person needs to understand what the journalist covers and how to approach them.

This tool helps PR and campaign teams understand how data stories, reports, studies, surveys, rankings, and similar artifacts moved across public web channels. It is built for pattern discovery and outreach strategy, not definitive origin claims.

Unlinked Mention Finder searches the web for everywhere your brand is named — reviews, news pieces, best-of lists, comparison pages, local guides — then opens every one of those pages and checks the HTML for a real link back to your site. The pages that name you and never linked are the list it hands you, ordered by which site is worth emailing first.

A Wikipedia article is one of the most trusted assets a brand can have — and one of the easiest places to waste money on doomed attempts. This tool answers the question properly. It searches the entity's real public record across web, news, books, and academic channels, reads the strongest sources in full, and judges each one the way a Wikipedia reviewer would: is it independent, is it reliable, is the coverage substantial? Then it delivers a complete engagement: a clear qualifies/borderline/not-yet verdict with a confidence level and an honest account of how thorough the search was; specific improvement proposals for an existing article, checked against what the article already says; the other Wikipedia pages where the entity could legitimately appear; a fully cited article draft built only from verified, quoted facts; and — when the answer is 'not yet' — a realistic plan for the coverage that would change it. Everything is research and recommendations for human review: it never edits or submits anything to Wikipedia.

When people ask AI a question in your space, the answer they get depends on which AI they asked. This tool asks them all — a dozen or more leading models, plus several searching the live web — and maps the collective answer: a consensus board showing how many AIs say each thing, the points where their memories genuinely contradict (with the live-web AIs settling checkable facts), and a collective answer written only from what most of them agree on. Debate mode goes further: the AIs argue it out in blind steel-man duels and three independent AI judges from other labs rule on the answer that survives scrutiny.

When someone asks ChatGPT, Claude, or Gemini about your company, what does it actually say — and how much of it is true? This tool asks a panel of leading frontier AI models a fixed set of neutral questions about you, extracts what each one believes, compares them against each other, and audits every belief against your own website and the open web. It surfaces the flat-out-wrong claims (a fabricated founder, an acquisition that never happened), the confident-but-unverifiable beliefs, the strangest things AI thinks, and the important facts AI simply doesn't know — each one quoted verbatim, with the real fact where the truth is known. Only things the evidence actually contradicts are called wrong.

Experts can carry a compiled knowledge wiki — synthesized practitioner articles plus the underlying sourced facts. This tool answers a natural-language question from that wiki in one step: it navigates the wiki's map, concept articles, and cross-links, recalls the most relevant sourced facts, and returns a single grounded answer that cites each specific statistic. It replaces the older list/read/search wiki tools with one ask.

Reads a business's reviews across Google, Trustpilot, Tripadvisor or Yelp. Three depths: an instant rating snapshot; the raw individual reviews; or a voice-of-customer analysis that mines the review text for the themes customers raise, what they love versus complain about, your reply-rate and recent-sentiment reads, and a ranked list of what to fix first.

Answer a question about the active client from everything we know — what they said/wrote (their emails/Slack/Drive, cited), what we've decided/noted, and their brand profile. Use for specific lookups ('what did they say about X', 'are we hitting the requirement', 'fee history', 'who owes what', 'what's overdue') and for the big-picture read ('how's the relationship with this client', 'anything I should worry about', 'what's the state of play'). Returns a cited answer, labels each fact's source, surfaces conflicts, and refuses when nothing's on file.

Cited Q&A over the company's own process-knowledge store ('How We Work'). Every answer cites machine-extracted practices from our own documents, each carrying evidence breadth (established/emerging/one-off), the date range its evidence comes from, and evolution notes when a practice changed over time. Confidential material never enters answers. Refuses honestly when nothing is on file.

Pick a competitor and it resolves who they really are, then pulls real, current data from across the web — new content and pages, backlinks and rankings, whether AI assistants name them, news and sentiment, social and ads, reviews, hiring, tech stack, even funding — and turns it into a clear you-vs-them story: where they're ahead, what they're investing in, and the specific moves to make in response. 'You' comes from your client profile, and it confirms everything (competitor, websites, depth, and the category it'll probe) before it spends a cent. Run it on your own brand instead for a single-subject health snapshot.

This is for editorial QA: verifying claims, assessing source support, and identifying unsupported, risky, vague, or overclaimed statements.

For a topic, this tool decomposes a standing cross-vertical corpus of high-performing content into its working parts — story shape, news hook, mechanism, unit of analysis, comparison basis, asset type, tone — and finds where your topic has an opening, surfacing fresh ideas grouped by signal: proven formats untapped here, what's working right now, angles getting ahead of the curve, and wildcards.

Give it one idea or a whole batch. It pressure-tests the ways an idea can fail — already covered, weak evidence, likely criticism, SEO saturation, methodological flaws, and audience objections — checking the checkable ones against live data (a full read of the top-5 Google pages for “already written,” recent news coverage, and real search volume/difficulty/competition) and judging the rest like a sharp editorial board. The result is advisory, never a veto: even a weak idea gets an honest score and a path to make it stronger.

For a topic, this tool reads what's actually ranking on Google and compares it against what podcasts, Reddit, Hacker News, and recent research are actively discussing, to surface the angles the ranking pages leave uncovered — your opening to out-cover the top results, get cited, or pitch first.

When you've done a stretch of work in one conversation — research, reports, data pulls, drafts — this reads back over everything that was produced and the discussion around it and writes a single grounded executive summary you can hand to a stakeholder or keep for the record. It works from the actual files and results, not just the chat, so the recap reflects what was really produced.

Finds a domain's highest-traffic pages and scrapes them into a CSV, for when you care about the pages that actually get the traffic. For a list of URLs or a whole site, Site Intelligence & Crawler is faster and free.

Give it a file you uploaded or a link, and it transcribes the audio and works out who is speaking. Every quote it returns is checked word-for-word against the recording, and every timestamp is real — it will leave a speaker unnamed rather than guess, and it will drop a quote rather than invent one. Podcasts and uploads get full speaker separation; platform videos use their captions, which are more accurate and far cheaper.

This is the general research tool when the team does not already have exact URLs and needs source material from the web.

This helps teams navigate available data assets before writing code. It can suggest datasets, columns, joins, and analysis paths from semantic metadata.

Hand it a messy data file (CSV/Excel/JSON) or a data URL and it surfaces the newsworthy, defensible angles — rankings, trends, correlations, and cross-table insights — with every number computed in SQL so the findings are provable, not hallucinated. Built for turning a survey export, metrics dump, or public dataset into sourced, ready-to-pitch story angles.

This is useful once a dataset is identified and the team wants to preview columns, dimensions, rankings, or values before deeper analysis.

Pulls live market data for listed companies, then frames it for research, PR, marketing and account work.

When you ask to watch / pin / attach / keep current a client document (a Google Docs or Google Sheets link, a file uploaded in this chat, or an asset already on the client) — and also when a client is active and you pastes a bare Google Docs or Sheets link that plausibly belongs with that client (a CM doc, guideline, brief, press list, do-not-contact list, tracker): a confirmation card with a relevance check is shown before anything is saved, so you decides. If they decline, read the doc and answer instead.

Give it a study or data-rich article and it produces a coordinated content package styled to your brand: a fact-checked social carousel (with PDF), a slideshow reel, kinetic typography videos, an explainer animatic, and an audiogram — with the full explainer render and an AI presenter available as one-click upgrades. Every piece is generated from verified claims and seeded so repeat runs look genuinely different.

This is the data-storytelling path: give it your stats (or wire them in from a research step) plus a few choices, and it produces a polished, on-brand infographic. Because Fractl's credibility depends on getting the numbers right, every result is independently verified by a panel of frontier models before you publish.

Two ways in. Give it a topic and it goes and researches it, deliberately hunting for material a normal deep-research pass misses, then tells you what it found and what that can become. Or hand it a link, a document or pasted text and it reads the whole thing and tells you honestly how much of it an episode can actually carry. Either way you choose, and it writes, fact-checks, revises and records the conversation.

This is for making a photo speak an audio clip, often using a generated or cloned-voice voiceover.

Paste a social URL and get its live numbers in seconds: views, likes, comments, followers. Or point it at an Instagram, TikTok or YouTube profile and ask for top posts: it pulls the recent feed and ranks the posts by engagement, with each post's full caption, date, metrics and link — the fastest way to find the post worth expanding into a newsletter feature or repurposing elsewhere.

For a topic or niche, Video Genome ingests the best-performing TikTok, Instagram Reels, and YouTube Shorts plus their most-engaged comments, then produces a two-tier deliverable: a punchy client content strategy (what to make and why) and a deep analyst reference with the evidence.

Give it a YouTube channel (link, @handle or name), a video link, a list of channels, or a topic. A channel comes back in full: subscribers, total views, how often it posts, the median views and the five best recent videos against it, engagement rate, the topics it covers and a one-paragraph read. A video comes back with its numbers, how it compares with the channel's recent median, and a digest of what the comments say. Channels can be compared in one table, and a topic search finds channels or videos with their numbers. Free official data, in seconds.

When a client asks 'what did you do for us this month?', this reads every conversation, deliverable, automated run, completed to-do and measured change on record for the period and writes it up in plain words for the client. Three shapes: a General update (everything, grouped the way the client would think about it), One topic (only the PR work, only AI visibility), or By area (Organic, Content, AI Visibility, PR, Technical, Links, Local). Every number and word is checked against the facts; the report lands in the client's deliverables, where you can edit it, share it through a review link, or download a branded PDF. Nothing is sent to the client automatically.

Tell the agent who and what — "tell Priya the Acme report is ready". It finds the person on your teams, shows you a card with their name, email and the exact text, and sends only when you approve. You can point at one file they can already open and the message carries a link to it.

Fans out across a dozen evidence channels: public datasets it analyzes itself (original findings that exist in no published article), community threads at scale (the receipts and lived patterns official averages hide), expert interview transcripts, computed public-data series, and a multi-angle published-web sweep that then audits its own coverage against what a competent deep-research pass would hold and sweeps the gaps. Every number is verified against the page it came from at deposit time. The output is a readable cited brief plus a persistent research file that Ask, the analyst workbook, and the Research-Backed Writer all build on.

Plans a set of table questions for the topic — each an inventory of named things of one kind with two to four attributes, or a figures table by year, region or sector — and, if you add your own questions, builds a table for each of those too. It searches with keyword queries, reads a few hundred pages, extracts every claim with the quote it came from, checks each cell against its page, and groups repeated reports under one row with every source kept. Three minutes later you have the tables, the evidence behind every cell, and a JSON the agent can analyse or answer questions from.

Consumes a research file built by the Original Research Sprint: reads all the collected evidence, drafts the whole piece in one sitting, then three independent editorial judges score and revise it (information gain vs a no-data baseline, cold-reader experience, evidence utilization) until it stops improving. Every number in the final piece is deterministically verified against the specific evidence it cites. The differentiation vs a 'deep research' run: deep research produces an unverified memo you still have to fact-check and rewrite before publishing; this produces the publishable piece itself — verified numbers, community receipts and original analysis no web summary contains, an attached reference appendix, and a provable before/after against what an AI writes with no data.