Content generation & optimization

When a scan finds a prompt you're losing, the usual fix is content. Stride drafts that content against the gap it found rather than from a blank page.

How it connects to your scans

Content generation is driven by scan evidence: the prompt you're missing, what the engine cited instead, and which of your existing pages came closest. That context goes into the draft.

Content Gap — measured diagnosis, not a queue

Content Gap (/dashboard/content/gaps) merges your tracked-prompt performance with the citations from your latest trusted, comparable full scan and shows a count-led gap for every tracked prompt: how many measured answers named you, how many cited your own site, and which competitor pages engines cited instead. Rows are ordered by competitor breadth. Open a row to review exact measurement frames, source URLs, and disclosures in its detail drawer. The drawer keeps every measured denominator visible — Stride never reports "not measured" as a fabricated zero. If a configured competitor has no tracked domain, Stride says so rather than guessing which pages it cited.

The overview defines Prompt coverage as measured prompts with at least one own-site citation, divided by all prompts measured in that trusted scan. It separates prompts into Own site cited, Mentioned only, and Absent, and reports how many indexed pages are available to Optimize Content. These are observed counts, not a synthesized content score.

Prompts with no usable full-scan evidence yet appear in a separate Track before you plan list — real candidates worth tracking, but not yet a measured gap; Content Gap stays unavailable for them until trusted measurement exists. Content Gap never claims a causal lift from a scan comparison, and it is a read-only diagnosis, not a work queue: everything already drafting, delivered, or being measured lives at Content Agents instead.

The detail drawer groups the prompt, coverage, scan date, type, and indexed page context in a fixed dark header. Below it, one vertical selector offers up to three deterministic next-step formats from the existing Content Agent catalog; only the selected format exposes its action and Evidence controls. A mapped page leads with Optimize; competitor evidence can surface Comparison; focused article and how-to formats fill the remaining choices. Selecting one seeds the owned prompt and format (plus the owned page for Optimize), and the receiving route re-validates every seed against the site. Nothing generates until you review the setup and choose Generate.

Knowledge Base — see exactly what Stride knows

Before scan evidence becomes a draft, Stride checks it against Knowledge Base (/dashboard/content/knowledge). The page separates three kinds of knowledge so evidence from one context cannot silently become proof in another:

• Your brand — browse the claims Stride may use about your product, grouped by fact type. These claims come from facts your team entered, sources your team supplied, or verified pages on your own site. This view also contains Stride's site understanding, Brand voice, and Personas.
• Competitors — browse claims backed only by evidence from the official domains configured for each competitor. Your uploads, pasted notes, and own-site pages never become competitor proof.
• Internal sources — manage the exact public pages, document files, and notes your team supplied. Stride versions each source and retrieves from the latest ready version automatically on every new Content Agent run.

The curated Brand and Competitor lists lead with the inferred statement Stride may use. Expand a statement to inspect its exact source, quote, confidence, observation date, and correction actions. In Internal sources, the table shows an extracted preview; expand a source's details when you need the parsed text. That keeps the main knowledge surface readable without hiding the underlying evidence.

Facts age out on a schedule—as fast as 30 days for pricing and longer for slower-moving claims—and are automatically held back from new drafts. You can inspect, correct, archive, or manually confirm an outdated record, but routine review is optional. When a draft's evidence gate identifies a gap, Action Workbench can re-read configured official competitor pages through its evidence-recovery step. Brand voice sets the tone and terminology every draft starts from. Personas are reusable reader profiles; editing one changes future drafts, never a draft already written.

Adding and managing sources

Sources are tenant-scoped and versioned. The library is built from three kinds of input:

1. Page URLs — paste one public URL per line. Stride fetches and processes only those exact pages; it does not crawl links or follow a site map.
2. Files — upload PDF, DOCX, TXT, Markdown, HTML, or CSV. These are the six core supported formats; other file types are rejected.
3. Notes — paste free-form text directly into a note source.

After you add a source, Stride queues it for processing, splits the ingested content into chunks, and stores the chunks with embeddings for retrieval. A successful refresh creates a new immutable version in knowledge_source_versions and advances the source's current version, so older draft outputs can still point to the exact version that grounded them. You can refresh a source to re-ingest it, or archive a source to remove it from active retrieval while keeping its history.

What sources are not

Knowledge Base sources do not turn Stride into a general web crawler or enterprise file system:

• Stride does not run OCR or extract text from scanned PDFs.
• Stride does not process archives (ZIP, RAR, etc.) or executable files.
• Stride does not connect to Google Drive, Notion, or other third-party document platforms.
• Stride does not recursively crawl websites from a seed URL.

A customer-provided source can support claims about your own product, but it can never stand in for verified competitor evidence. Claims about competitors still require the official, first-party, or measured evidence Stride uses elsewhere, and are tracked in the same claim ledger that Content Agents and the Workbench respect.

Content Agents, Create Content, and Optimize Content

Start a new piece from Create Content (/dashboard/content/create) — pick one of your active tracked prompts, then configure what Stride should draft: content type, length, an optional Persona and tone of voice, up to three sources cited in that prompt's own measured scan answers (never a fabricated rank — citation order is not evidence of an engine's own ranking), and a short brief for the type you picked. Stride generates the page title with the draft; you can refine it during review. Save as draft keeps an incomplete configuration without calling a model or using a content credit; only Generate spends one standard content credit.

To improve a page you already have, open Optimize Content (/dashboard/content/optimize). It opens your optimization portfolio first, not a setup form. Select Run optimization to choose an active tracked prompt and one of your indexed pages. Stride then edits only the sections that need to change to answer the prompt directly — the page's existing format, voice, links, and examples are preserved everywhere else. There is no content type or length to choose: the page's own shape decides that.

The portfolio groups saved work into Work, Tracking, and Evaluated. These are read-only views of the same Action Workbench and proof record: Work still needs setup, review, or implementation; Tracking means an applied change is waiting for comparable evidence; Evaluated means its evidence window has reached a result, not that Stride assumes the change produced lift. Select a prompt title to resume its saved Optimize editor or, after generation, the canonical Workbench where review, delivery, and proof continue.

Proposed section edits are reviewed one at a time and accepted or rejected; decisions are append-only, so a rejected edit stays rejected rather than silently reappearing. If a selected source changes before you copy a draft out, Stride blocks copy-out as stale until you recheck it. Each accepted revision becomes a new immutable version — earlier versions stay saved and selectable, and reopening one only changes what you're viewing.

Reviewing and editing a draft

Every draft opens in one editor — title, meta description, and body together, laid out to read like a finished page rather than a text box. Changes save automatically as you work; reload the page and your edits are exactly where you left them.

Alongside the draft, a GEO and SEO checklist lists the concrete requirements Stride checked for this version — things like naming your topic up front, covering the sections Stride planned for, and keeping your title and meta description in a healthy length range. Each shows as "X of Y requirements met," never a score out of 100, and only counts requirements that actually apply to this draft.

A Workflow view on the same draft shows how Stride produced this exact version — what it researched, how it planned the piece, and what it wrote — so you can see the reasoning behind what's in front of you. Older drafts saved before this detailed history existed say so plainly rather than inventing steps that didn't happen.

Stride runs a factual and intent check when it generates a draft. If you edit the title, description, or body afterward, that check is flagged as needing another look rather than shown as still passed — your edits always take priority, but Stride tells you when its own review is out of date.

Every configuration and run — draft, generating, in review, delivered, measuring, or verified — is listed at Content Agents (/dashboard/content/agents), whether it started from Create Content, Optimize Content, a Content Gap diagnosis, or a content recommendation in Actions. Content Agents is read-only: it never publishes or ships on its own, and review, delivery, and proof stay in the same Workbench Actions opens.

Where to work content

Whether you started from Create Content or from a content action in your Actions queue, the work happens in the same Workbench. The typical path:

1. Plan — confirm the objective, destination page or new-content route, and comparison scope when pricing is involved. If Stride finds an existing page covering the comparison, choose here whether to update it or create a distinct page; Create will not ask again.
2. Create — generate and edit a draft grounded in your scan evidence and site content; editorial gates block thin or unsupported claims
3. Deliver — publish to a connected Shopify site, copy into your CMS, or record the implementation
4. Measure — after a passed declared live check, compare later trusted, comparable full sweeps with the recorded target and baseline

Completed and current stages reopen their work panel. Future stages stay visible and unlock in order, so the journey doubles as navigation without letting work skip required decisions.

Pricing changes the evidence and page-ownership checks. When you include pricing, Stride carries your selected destination into a linked pricing comparison and verifies official pricing sources before drafting. If the pricing objective fits a different existing page, Plan explains the conflict and asks you to confirm that route instead of silently replacing your choice.

A generated draft or a recorded implementation is not proof that a change is live. When work is recorded, Stride links the exact generated output to the implementation event and canonical Action proof record. Advancement still requires a passed declared live check and qualifying target-specific comparable evidence.

This is separate from Community Signals, which helps you draft talking points for Reddit threads — Stride never auto-posts there.

Workflow modes (site setting)

Under Settings, choose how aggressively Stride moves from proposal to draft:

Workflow mode applies to content and optimization work in the Workbench — not to community outreach drafts, which always require you to copy and post manually.

Reviewing optimization proposals

When Actions has an optimization proposal, select Review proposal to see the evidence and questions it includes. Approve & generate drafts creates drafts only; it never publishes a change. Dismiss proposal leaves each question tracked, so a later scan can surface a new proposal.

Citation-change signals are separate from your work queue. Select Not now to hide one until the next comparable scan without removing the question from tracking.

Monthly limits

Content generation draws on a monthly allowance set by your plan. Additional credits are available as add-on packs — see Plans & limits.

Revise a saved draft

Use Revise draft when a saved draft needs a focused change. Before you submit, Stride shows that the revision uses one content credit. If Stride cannot save a usable revision because generation fails or a quality check blocks it, the credit is returned.

Each successful revision becomes a new saved version. Use Version 1, Version 2, and later version controls to compare or reopen earlier work. Previous versions remain saved, and choosing an older version only changes what you are viewing — it does not delete newer work.

A revision keeps the destination and content approach you confirmed in Plan. It cannot silently switch either choice or publish the draft. Change planning decisions in Plan, then use Deliver when you are ready to publish or record the implementation.

Editing and publishing

Drafts are a starting point, not finished copy. Review everything before it goes live: AI-drafted content that ships unedited tends to read like AI-drafted content, and that works against you with both readers and engines.