Changelog

What's new in Wavefront: features, integrations, and fixes as they ship.

August 2026 - A tool catalog your agent can actually use

Essential tool profile for connected agents: Cursor, Claude, and other MCP

clients now see a curated ~40-tool catalog covering the whole core loop (orient, plan, brief, draft, validate, images, publish, measure) instead of 220+ tools. Small catalogs make coding agents dramatically more reliable at picking the right tool.

Full catalog one flag away: append ?tools=all to the server URL to list

everything (Link Network, visibility battery, beliefs, webhooks, admin). The flag only changes the listing; any tool your role allows stays callable by name on either profile, so existing agent workflows keep working unchanged.

The tool catalog now marks essential tools, and

Settings shows the flag next to the connection config.

August 2026 - Portfolio: all your workspaces on one screen

New Portfolio page for anyone who belongs to more than one workspace (agencies,

consultants, multi-brand operators): one card per workspace with what's waiting on you, articles published this month, the AI visibility score, and credits used. One click opens the workspace. Find it above the workspace switcher, or search "portfolio" in ⌘K.

Every new workspace starts with its strategy in place. Signup now seeds the

default editorial strategy, so the Plan page is never empty on day one.

Failure messages you can act on: run and job failures shown on Home, Activity,

and run detail now explain what happened in plain language ("another run was already in progress") instead of internal runner strings.

Clearer autopilot settings: "Posts per run" is now "Articles per cycle," with

an explanation of how the hourly cycle and your calendar dates interact.

July 2026 - The belief ledger: what AI engines actually say about you

Beliefs tab on Grow: every AI visibility scan now feeds a ledger of what each

answer engine currently asserts about your brand, per question and keyword. Wrong statements (the engine quoting incorrect facts about you) surface first, verbatim.

Positioning targets: declare where you want to be mentioned or recommended;

the gap view shows exactly where each engine stands against that target (wrong, not observed, below target, stale, or achieved). Brand-description accuracy is checked automatically, no setup needed.

Agent access: six new MCP tools (list_beliefs, derive_beliefs,

list_belief_targets, set_belief_target, remove_belief_target, list_belief_gaps) so your own agent can read and steer the same ledger.

July 2026 - Fix: scheduled posts can no longer get stuck "writing"

Autopilot runs now record which calendar entry they were dispatched for. When the

run finishes, the entry is linked to the draft it produced even if the agent chose a different title or slug; if the run fails or writes nothing, the entry returns to the schedule for retry instead of showing "writing" forever.

July 2026 - Visibility OS: the last mile

The remaining plan-spec items shipped, closing the gap between the plan of record and the product.

Declared crawler policy with drift detection: training/search/retrieval choices default to "undecided" (report + nag); once declared, the audit verifies robots.txt matches your intent in either direction, catching the "we thought we blocked it" false-protection case.
Semantic neighborhood: one command drafts the buyer-language terms your content should co-occur with; you approve them, coverage over the catalog is measured deterministically, uncovered terms become topic suggestions, and matching keywords get a +10 opportunity bonus.
Organization schema generator: generate_org_schema builds the JSON-LD block (canonical description + sameAs to every registered profile); the audit verifies installation and flags incomplete sameAs. Plus the community-squat check when engines cite community sources in your category.
Hedge sources: every hedged answer now shows the citations behind it. Named sources = outreach task; no sources = content task. The finding and the Trust tab both say which.
Claim credibility measured: substantiated statistics across recent articles feed the trust score, using the same structural check draft QC enforces.
Audit-time fan-out: full audits map your top-5 keywords' fan-out automatically (cached for a quarter), and generate_draft requires uncovered sub-queries as sections. mark_reviewed stamps a freshness review without a rewrite; high-traffic and AI-cited pages now age out at 180 days.
Packaging enforced end to end: competitor, surface, and business-critical-prompt caps per plan; Free's one lifetime full audit (then honest lite runs); Citation Core from Starter; a platform-wide daily budget cap on the public /audit teaser; white-label shared reports on Scale.
Grow polish: layer owner chips ("who owns this?" with the course's team mapping), cross-engine agreement on the consensus row, visibility suggestion chips in the assistant, and specialist-over-generalist concentration nudges in topic suggestions.

July 2026 - The visibility loop closes: roadmaps, board packs, and a free public check

Live 90-day roadmap: one command turns your open findings into a phased 30/60/90 plan with layer owners, saved as a Library doc that regenerates on demand and auto-ticks items the next audit no longer detects. The living alternative to a deck that's stale the week after it's presented.
Executive overview export: a CMO-ready board pack ("The Shift" framing, the five-layer scorecard with owners, top findings, the competitor citation gap, and a methodology appendix that says exactly how the numbers are made). Frozen snapshot, revocable public share link on Pro.
Scheduled audits by plan: weekly on Pro and Scale (staggered per workspace), monthly on Starter, and every plan keeps unlimited manual runs. Battery size is plan-packaged too (8 / 20 / 30 prompts) and the pricing page renders straight from the code, so what we sell can't drift from what runs.
Free public visibility check at /audit: paste any URL, get a 0-100 technical score and the top findings in seconds. No signup, no email, no LLM: deterministic checks only, results kept 7 days, private addresses refused. The honest teaser for the full audit.
Docs: the full score methodology and per-layer guidance is now public at /docs/ai-visibility. Your weekly digest carries the scorecard line, and Home shows the score next to this month's output.

July 2026 - Discoverability, deep: and can an agent actually use your site?

The Discoverability and Agentic layers grew from summary scores into working surfaces.

Crawler function matrix: training, search-index, and live-retrieval crawlers are three separate decisions with separate user-agent strings. The new Discoverability tab shows exactly what blocking each one does, flags deprecated crawler names that silently do nothing (a robots.txt naming anthropic-ai protects nobody), and set_crawler_policy generates the robots.txt block for your deliberate choice. Googlebot is never auto-blocked (it's also web search).
Extractability rules: four new QC rules teach and enforce how AI retrieval reads passages: answer first (44.2% of citations come from the first third), self-contained sections, descriptive headings, and front-loaded evidence. Drafts are written and validated against them; the audit samples your published articles with the same checks.
Query fan-out mapping: AI silently expands every query into sub-queries with zero measurable search volume. map_query_fanout shows them and which ones your catalog covers; planning an article now maps its fan-out automatically so drafts close the gaps.
Freshness got type-aware: statistics and news pages age out at 90 days, evergreen at 12 months.
Metadata checks: title/meta-description quality, H1-title agreement, Open Graph consistency, and sitemap health joined the GEO audit.
AI referral traffic: sessions from chatgpt.com, perplexity.ai, gemini, copilot, and claude.ai referrers, snapshotted nightly from GA4. Small numbers, but AI referrals convert at ~4.4x organic; the trend is the point.
Agent-readiness audit (14 checks): can an AI agent complete a task with your brand? HTTPS, agent-bot access, link semantics, actionable conversion points, labeled forms, policy/contact/pricing discoverability, copyright hygiene, and your catalog's task-intent mix, on the new Agentic tab.
Catalog intent classification: every published entry classified (informational → task) in nightly batches; the agentic score reads the mix.
Agent card + protocol readiness: generate a machine-readable identity card from your canonical description, and get the honest MCP/ACP/UCP/AP2/A2A/WebMCP/NLWeb checklist (most are "watch" unless you transact online; your workspace already speaks MCP). Plus a seeded experiment tracker: run tactics as experiments, not beliefs.
Engine consensus on the score card: per-engine mention rates side by side, because engines cite from disjoint corpora and divergence explains score movement.

July 2026 - The Citation Core: who AI actually cites in your category

Citation ledger: every source a live answer engine cites in your scans is now recorded (the proxy panel writes nothing, honestly: it has no real sources). Data accrues from your first Perplexity scan.
Citation Core: a quarterly rollup of that ledger into the small set of domains AI actually trusts in your category, with category, cross-engine consistency, your presence vs competitors' on each, and a frequency-by-reachability priority grid (1 = quick win, 4 = honestly not worth chasing; encyclopedias and government sources are graded unrealistic instead of pretending).
The finding leadership responds to: "this core domain cites your competitors but not you" - concrete, fixable, and routed into your audit findings, along with the Wikipedia-gap note and quarter-over-quarter core movement (cores shift fast; chasing a fading source wastes outreach).
Link Network cross-reference: when a core domain is a Link Network member, getting cited there is one placement request away instead of a cold outreach project. Flagged right in the table.
Original-asset suggester: proposes the data reports, surveys, tools, and benchmark assets AI would cite because nobody else has the information, straight into your Review Inbox.
New Authority tab on Grow; for agents: build_citation_core, get_citation_core, suggest_original_assets.

July 2026 - Clarity and Trust: what AI says about you, verbatim

The visibility score grew its Clarity and Trust layers into full working surfaces.

Canonical brand description: the signable source of truth (description, category, value proposition, target customer, and the "what you are NOT" list). The onboarding wizard drafts it from your site for you to edit until you'd sign it; everything below judges against it, never against a guess.
The prompt battery: typed probes seeded from the AI Search OS course's own templates - brand questions (including "are there any complaints or issues?", the look-alike disambiguation probe, and "can I book, buy, or sign up through you?"), competitor comparisons, and best-for questions. They run first in every AI-visibility scan.
Wrong vs missing: brand-prompt answers are diagnosed against the canonical. Wrong claims mean third-party sources contradict you (fix the sources); missing facts aren't in anything AI can retrieve (publish them). The Clarity tab shows every answer verbatim, per engine.
Hedging is now measured: comparison answers classify as endorsed, hedged, not recommended, or absent, with the qualifying language quoted verbatim. The Trust tab shows what AI's hesitation actually sounds like, and hedge direction feeds the trust score as its leading indicator.
Business-critical alerts: flag the prompts that matter to revenue; if one stays non-endorsed for two consecutive scans after being endorsed, you get a visibility.regression webhook and a digest line. Single-scan flips never alert (AI answers are volatile).
Owned-surface drift scanning: register the places that describe your brand (homepage, LinkedIn, Crunchbase, review profiles). Monthly scans judge each against the canonical and hand you a copy-ready corrected paragraph on drift. Login-walled surfaces take a manual "I verified this matches" attest, re-nagged every 90 days. Review profiles get their rating and review count extracted automatically.
Review-request kit: one command generates ask copy timed to customer wins plus the context questions that make reviews quotable by AI (honestly labeled: the timing is a conversion best practice, the AI-citation upside is a bet).
For agents: draft_canonical_brand, list/set/remove_battery_prompt, seed_battery_prompts, get_battery_answers, list/set/remove/attest_surface, run_surface_scan, generate_review_kit.

July 2026 - Your AI visibility, scored

Wavefront now audits how visible your brand is to AI search engines and scores it 0-100 across the four layers that decide whether ChatGPT, Perplexity, and Google's AI answers mention, recommend, or act on you: Discoverability, Clarity, Authority, and Trust, plus Agentic readiness.

One button on Grow → Overview runs the Brand Visibility Audit. It reuses the signals Wavefront already collects (AI-crawler access, server-rendered content checks, your weekly AI-visibility scans, publish freshness) and shows exactly which components fed each layer's score, including the ones with no data yet and what to connect to light them up.
Findings, not homework: every issue the audit finds carries evidence, a severity, an honesty grade (best practice / bet / watch, mirroring how strong the research behind it is), and a fix route. Content-fixable findings land in your Review Inbox as proposals the assistant can write; the rest tell you which team owns the fix.
Findings close themselves: re-run the audit after a fix and anything the fresh audit no longer detects resolves automatically.
Layer owners: assign who owns each layer (Clarity usually belongs to product marketing, Authority to PR, Trust to customer success) so the cross-functional work has names on it.
New workspaces get a lite audit automatically at the end of onboarding: a score and your first findings within minutes of signup.
For agents: run_visibility_audit, get_visibility_report, list_findings, dismiss_finding, set_layer_owner over MCP, and the in-app assistant can read the report and findings directly.

July 2026 - A simpler dashboard: seven places, one review queue

The dashboard collapsed from fourteen destinations to seven, built around one question: what needs you?

Home now opens with a single Needs your review queue: article proposals, finished drafts awaiting your OK, paused assistant actions, and link-placement requests, all with the same Approve verb. A setup checklist replaces the scattered nudges.
Plan merges the strategy hub and the editorial calendar: one calendar, with strategy chips that filter it and open the selected strategy's panel.
Content merges Drafts and Published into one lifecycle list (Writing → Needs review → Ready → Published).
Activity merges run history, pipeline jobs, and the audit log into one timeline, so "what happened?" has one answer.
Settings absorbs Team and Billing as tabs.
New workspaces start writing by themselves: autopilot is on in review-only mode from day one; drafts land in your queue and nothing publishes without your approval.
Fix with AI on every draft: one click dispatches an agent run that repairs failing QC checks.
Search (⌘K) now finds calendar entries, published articles, and destinations ("go to billing"), and every old URL redirects to its new home.

July 2026 - Introducing Wavefront

Wavefront is now open for self-serve signup: an AI content engine that plans your SEO strategy, writes and QC-checks every article against your brand rules, publishes to your CMS, and reads your Search Console and Analytics data to plan the next move. Every change lands as a reviewable diff you can revert.

Launch surface:

The loop: strategy hub, editorial calendar, drafts with QC validation, publishing, and the weekly Growth Review that turns your analytics into evidenced proposals.
Auditability: per-run changesets with file-level and whole-run revert, run history with token and cost tracking, and a versioned audit log.
10 CMS integrations: Contentful, WordPress, Webflow, Ghost, Shopify, HubSpot, Sanity, Strapi, Markdown + Git, and a signed custom webhook, plus token-gated RSS/JSON feeds and a read-only content API.
Google connections: read-only Search Console and GA4 for the growth loop.
Bring your own agent: the full platform exposed as a remote MCP server (210 tools across 13 modules at launch) for Cursor, Claude Code, and Claude Desktop, on Pro and Scale plans.
Plans: Free (500 credits, no credit card), Starter $49, Pro $199, Scale $799, with metered credits at $0.01 and a spend ceiling you control.
This site: new marketing pages, documentation hub, FAQ, and expanded Terms and Privacy.