06-reference/research

goodnotes generated visuals canvas

2026-10-08·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)

Goodnotes does put generated objects on the canvas, behind a confirm-then-Insert step and a Pro paywall, and the sharper threat has moved to the connector shelf

The question

"What did Goodnotes' September 2025 'generate visuals directly in your canvas' release actually ship, and does the incumbent already place open-ended generated content on the page?"

Context: this is the unresolved fetch gap from [[2026-07-27-pen-first-ai-canvas-competitive-landscape]], which scored the Goodnotes threat on a press-release claim it could not retrieve (BusinessWire timed out, support.goodnotes.com returned 403). The backlog entry named the asymmetry correctly: if Goodnotes already writes open-ended generated content onto the page, a CROWDED verdict understates the threat, and that is the error direction that costs a sprint. The decision this now informs is Scribble Works, not ink.

What we already know (from the vault)

What the web says

Both failed fetches are now resolved, and the two workarounds both earned their keep. BusinessWire still returns 403 to curl with a browser User-Agent and was not retried, but the identical release sits on Goodnotes' own newsroom and returned HTTP 200 to that same curl call. support.goodnotes.com, which 403'd WebFetch in July, served cleanly to Playwright with no bot wall at all, which unlocked the ten-article Goodnotes AI help section that settles this question. Two further surfaces proved richer than the marketing pages: a maintained /llms.txt, and a /for-ai-assistants page stamped "Last Verified: September 2026." One trap for reuse: .md suffixes on goodnotes.com paths return the 404 page with HTTP 200, so byte-size comparison against a known 404 is mandatory. One target remains closed: the Goodnotes AI FAQ article is behind an end-user sign-in redirect, not a paywall, and was not retried.

What the September 23, 2025 release actually shipped. Per the vendor release: Goodnotes Essential and Goodnotes Pro embedded with Goodnotes AI; Whiteboards as an infinite canvas; Text Documents; and real-time multi-user editing across all three. Claimed scale: "over 25 million monthly active users." The sentence the parent brief could not verify exists verbatim, as one of four illustrative examples rather than a named feature: "Generate visuals from text or handwriting, including sketches, diagrams, and charts, directly in your canvas." The matching launch note is blunter: "Meet your new thinking partner that works directly in your notes."

Yes, the incumbent places generated content on the page, and the feature has a name, a toggle and a price. The help center draws the line at Create mode, not at the feature level. Create mode on: "AI assistant will create content and actively edit your document", with the Pro Tip "ideal when you want Goodnotes AI to take action and make changes directly to your document." Create mode off: "simply provide information", for use "without modifying the document." Create mode locked: "not available on your plan." The vendor's own framing is "co-create with AI on the canvas." The old Generate Diagram feature "has found a new home in Create mode", covers "flowcharts, mind maps, timelines and more", and carries the one strong editability claim in the corpus: "your diagrams will be fully editable once they're in your notes."

Three qualifications turn that yes into a bounded yes, and all three are the vendor's own words. First, placement is a confirmed two-step insert, not an autonomous drop: "Once you're happy with the created content, tap Insert to add it to your page. If you'd prefer to start over, tap Discard", then "you can move and place it anywhere on the page, giving you total control." The launch notes say "preview and modify any AI suggestions before applying." Second, the assistant itself lives off-canvas in a Floating, Sidebar or Window display, and at least one AI output never reaches the page at all: generated document outlines are confirmed with "Insert in Sidebar." Third, the documented output set is bounded to diagrams, images, templates, visual summaries onto a new page, quizzes and study sets, with explicit limits: "when editing content, Goodnotes AI typically creates new versions rather than modifying existing text directly", styling is "limited to those in your document editor", and it "cannot directly handle file operations." Nothing anywhere describes spatial addressing, free-form placement, or multi-step agentic behavior.

Editability postdated the press release, and covers diagrams only. The Oct-Dec 2025 roundup (published 2026-01-09) lists, typo included: "Editable AI-generated diagrams are now editable, and use improved algorithms to minimize overlapping elements." For generated images the docs claim only that you can "move and place it anywhere on the page", which is consistent with an inserted image object rather than a native editable one. No article claims generated text arrives as editable ink strokes.

The handwriting-render floor is real but narrow, and the new feature moves away from it. Spellcheck does render in your own hand: "Goodnotes will use AI to generate and match your handwriting style based on the handwriting on the page you are working on", so "pages with more information will produce more accurate results." It is on-device, costs no credits, is available on every plan including Free, and covers nine languages, but it is correction of an existing word, not generation of new content. The Q3 2026 headliner Restyle goes the other way: it "uses Goodnotes-provided handwriting styles", "custom handwriting styles are not currently supported", and it "does not alter drawings, diagrams, and other non-text strokes."

Gating is materially tighter than the pricing page suggests, and credits finally have absolute numbers. Per the help center's plan comparison, AI on Notebooks and Whiteboards - the two canvas surfaces - requires Pro or AI Pass. Essential gets none of it, and Essential's own line reads "does not include Create Mode, Meeting AI, or Edit AI." Pro's line: "Create Mode (AI content generation) for Text Documents, Notebooks, and Whiteboards; Image Generation; Edit AI." Monthly AI credits: Free 10, Essential 525, Pro 525, AI Pass +6,300, replenished on the 1st, where "525 credits typically powers about 3 hours" and "6,300 credits can fuel approximately 30 hours." So the real entry price for on-canvas AI generation is $35.99/yr, and the heavy-use price is roughly $155/yr once AI Pass is attached.

Pricing re-verified today and unchanged in thirteen months. From goodnotes.com/pricing on 2026-10-08: Free $0, Essential $11.99/yr, Pro $35.99/yr, Teams $120/seat/yr, Enterprise custom, Education $0, AI Pass +$9.99/month. All base plans are annual-only; AI Pass is the only monthly charge and is not available on Free. Minor conflict worth carrying: the live pricing page says $9.99/month twice, while /pricing.md and the help center both say $10/month. There are no overage charges; you upgrade instead.

Goodnotes took the assistant-connector shelf in April 2026, for both ChatGPT and Claude. The help article "Use Goodnotes in ChatGPT and Claude" (index-dated 2026-04-13) says: "Use either AI service to create text documents, whiteboards, and diagrams, then import the result as a new document in your Goodnotes library." Diagrams come as mindmaps, timelines and flowcharts in Classic, Gray or Line styles. Critically for channel economics, it "does not use your Goodnotes AI credits" and is "available on all Goodnotes plans, including the Free plan." Its stated limit: "the integration creates new documents only. It cannot edit or analyze documents already in your Goodnotes library."

Scale is up sharply, and the only engine ever named is Wolfram. App Store lookup API, 2026-10-08: 4.72 stars across 460,106 ratings, version 7.1.27 released 2026-10-05, against 381,000+ in July 2026. No marketing page names a model vendor; the one attribution anywhere is Goodnotes AI for Math, which "leverages Wolfram|Alpha LLM API." Stated posture: "Is my data used to train the AI? No."

Convergences and contradictions

Synthesis for RDCO

The verdict should split, and the split is the whole value of this brief. On the ink-shaped product axis, the Goodnotes component tightens, modestly and specifically: Create Mode is a real named SKU that acts on the document, diagrams land as editable native objects, and the price of entry is a $35.99/yr consumer subscription against 460,000 App Store ratings. That is a better-armed incumbent than the parent brief scored. It does not tighten to the degree the backlog entry feared, for three reasons the vendor itself supplies: generation is a preview-then-Insert transaction rather than the agent placing content where you pointed, the output set is enumerated object types rather than arbitrary content, and the assistant lives in a floating panel, sidebar or window with the canvas as a destination rather than as the conversation. The two mechanics the parent brief identified as ink's durable asset survive. Handwriting-matched rendering exists at Goodnotes only as spellcheck correcting a word you already wrote, and the new Restyle feature explicitly refuses custom handwriting styles.

For Scribble Works, Goodnotes should be scored as a non-competitor on product, and the evidence is unusually clean because it is an absence in the vendor's own documentation. Help-center search returns zero results for "printable", one result for "worksheet" which tells you to import a worksheet PDF for math solving, and one result for "kids" which is about signing in with a child's Apple ID. There is no documented path from any AI output to a printed page. The buyer is a knowledge worker, a student or a school procurement officer; the kid-facing motion runs through ClassLink and Eduplaces roster sync, meaning student accounts in a school tenant on Apple-only hardware, which is exactly what SW-R18 forbids. A large AI-adjacent incumbent is not automatically a threat, and this one is in a different business. The CROWDED verdict, read as a statement about Scribble Works, should loosen to near-zero on the Goodnotes product axis.

The live risk is the connector shelf, and it is now five months old rather than hypothetical. Scribble Works' claim is "parent asks their agent for a Halloween hunt in Spanish for a 4-year-old; a PDF lands," and it remains an unbuilt plan in scribble-works#333. The three engine gaps that note names (agent-safe auth, a catalog read API or MCP endpoint, an agent lane on the rate-limited generative rails) are now competing for shelf placement against an incumbent that cleared all three in April. Two things make that less bleak than it sounds. The Goodnotes connector is deliberately thin - "creates new documents only, it cannot edit or analyze documents already in your library" - so it is a generation on-ramp, not a platform. And its output is a document, a whiteboard or a diagram, which is undifferentiated from what the host assistant already does, whereas "printable kids' activity, graded against a rubric, PDF out" is not. The honest read is that #333's sequencing got more load-bearing than its build cost, and that the thing which actually gates it is the accept-rate number in [[2026-10-01-accepted-spec-v0]] that does not exist yet.

Two smaller takeaways worth banking. First, the pricing anchor is now verified stable for thirteen months and much better specified than before: $35.99/yr for on-canvas AI, +$9.99/mo for roughly twelve times the credits, 525 credits as "about 3 hours," no overages, upgrade-only. For a consumer AI product at this scale that is a real signal about both ceiling and packaging, and it is independent of pen-first positioning. Second, Goodnotes Education's "pre-approve AI-powered hints that students reveal one at a time" is a pattern Scribble Works can lift without touching SW-R18: an adult approves the scaffolding in advance, the child consumes it progressively, no kid account required. On paper that is a fold-under or scratch-off hint strip, and it belongs beside the informational-praise scripts already specified in [[2026-09-19-mentava-motivation-mechanics-paper-translation-scribble-works]].

Why this is in the vault

It closes the one flagged unverified claim in [[2026-07-27-pen-first-ai-canvas-competitive-landscape]] against the vendor's own help center, so that brief's Goodnotes row can be restated as "Pro-gated Create Mode that acts on the document via confirmed Insert, diagrams editable, bounded output set" rather than "unverified, possibly an understatement." It also hands RayDataCo/scribble-works#333 a dated fact the issue does not contain: a 25-million-MAU incumbent has been on the ChatGPT and Claude connector shelf since April 2026, free on every plan, which is the sequencing argument for or against building the Scribble Works connector lane now.

Open follow-ups

Related

Sources

Vault

Web — Goodnotes help center, reached 2026-10-08 via Playwright after curl/WebFetch 403s. Dates are help-center search-index dates; no article page displays a last-updated line.

Web — Goodnotes first-party pages, fetched 2026-10-08 via curl with a browser User-Agent

Flagged — not usable as fact