The short version
- Dice generates 30 possible topics internally, checks live results, and recommends the strongest 15 rather than asking you to publish everything it can think of.
- The final plan is split into five Start now articles, five Build next articles, and five Later articles so the output is a sequence, not another backlog.
- Quick Wins use existing discovery and enrichment evidence to find relevant ranking articles that are thin, basic, dated, or otherwise beatable.
- Any recommendation can become a project with its search evidence, a starter blueprint, an improved brief, first-hand inputs, and a review-first working draft.
- Scans only run when you start them, reuse cached pages, and enforce a Firecrawl credit ceiling. Nothing is published automatically.
A blank content calendar looks simple until you try to fill it. A language model can produce fifty plausible titles in seconds, but plausibility is not the same thing as opportunity. It has not necessarily checked whether people search for the topic in that form, what currently ranks, whether your site already covers it, or what a new article could contribute that the existing results do not.
That is the problem the Content workspace in Dice is designed to solve. It creates a small, ordered publishing baseline from the site's actual market profile and live search evidence. The first output is called the Starter 15 because fifteen is enough to establish useful coverage without handing a new site a forty-article obligation before it publishes the first one.
The goal is not to generate the largest possible backlog. It is to identify the smallest useful publishing sequence and show the evidence behind it.
What happens when you build a Starter 15
The scan begins with the selected site's profile: what the company does, who it serves, its topics and keywords, and the competitors already attached to the profile. Dice uses that context to plan a broad mix of educational guides, comparisons, alternatives, problem-led articles, templates, and purchase-support content.
- 01
Plan 30 possible topics
The first pass deliberately starts wider than the final plan. Every candidate includes a proposed title, a realistic target query, its topic cluster, search intent, funnel stage, page type, destination, suggested slug, and a reason it belongs on this site.
- 02
Check the live results
Each target query is searched without scraping every result. The scan records the result URLs, titles, descriptions, and positions, then selectively reads the leading pages where the most useful evidence is likely to be.
- 03
Look for a real opening
The evidence is evaluated for customer relevance, clarity of intent, strength of the current results, freshness, differentiation potential, and how much confidence the observed pages justify. No search volume or traffic number is invented when the provider did not supply one.
- 04
Select and order 15
The strongest unique candidates become the visible plan. Each retains the ranking evidence and the reasoning that put it ahead of the other ideas, so its score can be inspected instead of accepted on faith.
- 05
Check for an existing customer page
Before calling an idea new, the scan checks whether the customer's own domain already has a clearly matching page. When it does, the recommendation becomes a refresh instead of a duplicate article.
The search stage follows the same principle as discovery query planning: use several deliberate routes into a market, retain how a page was found, and make expensive reading selective rather than automatic.
Why the plan is five, five, and five
A ranked list still leaves one practical question: where do we stop? The Starter 15 answers it by turning the ranking into three explicit publishing horizons.
| Bucket | Articles | How to use it |
|---|---|---|
| Start now | Ranks 1–5 | The clearest combination of relevance, visible search opportunity, evidence, and differentiation. Begin here. |
| Build next | Ranks 6–10 | Important supporting coverage once the first group is moving through research and production. |
| Later | Ranks 11–15 | Useful expansion topics that belong in the library but do not need to delay the stronger opening set. |
The buckets do not predict publication dates. They make the recommendation actionable. A two-person team might work through Start now over two months; a larger team might commission the first ten together. The ordering survives either workflow.
What every recommendation contains
The title is the least interesting field. Opening a recommendation reveals the plan and the evidence that make the title useful.
| Part | What it answers |
|---|---|
| Target query and intent | What the reader is trying to accomplish, not merely which phrase appears in a heading. |
| Priority score | How the idea compares with the other candidates in this scan. |
| New or refresh | Whether the customer needs a new page or should improve a matching page already on its domain. |
| Recommended angle | The specific version of the topic that could be more useful than the pages already ranking. |
| Reader outcome | What a visitor should be able to understand, decide, or do after reading. |
| Questions and outline | The minimum shape of a complete answer before a writer begins drafting. |
| Differentiators | Concrete ways to improve on the current results instead of paraphrasing them. |
| Evidence still needed | Facts, primary sources, product proof, and first-hand examples that cannot honestly be generated. |
| Ranking evidence | The live result pages that informed the recommendation. |
Quick Wins: the ranking pages that look beatable
The second view in the Content workspace begins from a different direction. Starter 15 asks which subjects the customer should cover. Quick Wins asks whether Dicee has already encountered a relevant ranking article whose execution leaves a credible opening for something better.
Quick Wins reuse pages that have already passed through discovery and page enrichment. That matters because the decision is based on more than word count. It can consider the query and rank that surfaced the page, content relevance, editorial quality, depth, freshness, placement suitability, and whether competitor evidence suggests room to differentiate.
- A page must have ranking proof in retained discovery evidence. Dicee does not label an arbitrary weak article a Quick Win merely because it looks bad.
- It must still be relevant to the customer's market. Outranking an unrelated page would not build a useful content library.
- Weakness can come from several signals: thin or low-quality analysis, shallow coverage, missing decision criteria, an old publication or modification date, or a visible competitor gap.
- Spam and unsupported-language pages are excluded. A low-quality result is only useful as evidence when Dicee can judge it reliably.
A new site may have an empty Quick Wins view. That does not mean there are no weak pages on the web; it means the workspace does not yet have enough enriched discovery evidence to make that claim. Running and enriching opportunity discovery gives this view the source material it needs.
From recommendation to content project
A recommendation becomes useful work when someone decides to act on it. Selecting Start project creates a focused workspace rather than sending the idea straight to a one-click article generator.
- 01
Evidence
The project keeps the selection reason, score breakdown, result URLs, and the page to beat or refresh. A writer or editor can inspect the source of the decision before accepting the angle.
- 02
Starter brief
The recommendation's blueprint becomes an immediate brief without another generation request. It contains the angle, reader outcome, questions, outline, differentiation ideas, required evidence, and call to action.
- 03
Improved brief
When requested, the existing OpenAI integration expands the blueprint into a writer-ready brief with a section-by-section objective and key points. It continues to separate known evidence from research still required.
- 04
First-hand inputs
The customer can add product facts, internal examples, opinions, processes, test results, and customer proof. These are the details that make the finished article something a competitor cannot reproduce with the same prompt.
- 05
Working draft
A draft can be generated from the brief, retained sources, and customer inputs. Unsupported experience becomes a clear customer-input placeholder instead of a fabricated anecdote. The result is saved for review and is never published automatically.
The workflow deliberately separates choosing an opportunity, planning the answer, supplying first-hand proof, and writing. Compressing those into one prompt is faster right up until someone has to trust the result.
How cost is controlled
A research feature can become more expensive than the content it recommends if every intermediate possibility triggers a full scrape and a full model pass. The Starter 15 therefore treats cost as part of the workflow rather than an account-page surprise.
- A scan only runs when someone clicks Build Starter 15 or Refresh Starter 15. There is no automatic schedule.
- Search begins with metadata. Only the leading evidence pages are read in full.
- Fresh page content already stored by discovery is reused instead of scraped again.
- The planned Firecrawl maximum is 150 credits, with a hard task ceiling of 160 estimated credits.
- Content evidence uses basic page fetching and disables PDF parsing, avoiding the variable costs of enhanced fetching and per-page PDF processing.
- Whole paid strategy and writing jobs do not automatically replay after failure and silently generate the same output twice.
- The existing OpenAI key is reused. There is no second Dice-specific AI credential, though normal OpenAI usage still belongs to the connected account.
The same principle appears elsewhere in the product: scanning in bounded batches keeps expensive work explicit and recoverable instead of pretending provider usage is free.
What the feature does not promise
A basic article ranking today is evidence of an opening, not a guarantee that a replacement will outrank it. Rankings also depend on the authority and technical health of the site, links, intent match, publication quality, competition, and how the result set changes after publication.
- Dice does not invent exact search volume, traffic, or keyword difficulty when those metrics were not supplied.
- A high score means the idea is stronger relative to the other observed candidates. It is not a forecast of a particular ranking position.
- A working draft is not a publish-ready claim that research is complete. It is the point where editorial review begins.
- Quick Wins are evidence-bounded. If the workspace has not discovered and enriched a ranking page, the feature does not pretend it has analyzed it.
- Nothing is sent to a CMS or made public automatically.
A practical way to use the workspace
- 01
Confirm the site profile
Good recommendations begin with an accurate description, audience, topics, keywords, and competitor set. Correct the profile before paying to research the wrong market.
- 02
Build the Starter 15
Read the first five as a group. Check whether they cover distinct jobs for the reader rather than fifteen variations of one keyword.
- 03
Inspect the evidence
Open the result pages and challenge the proposed angle. Evidence and manual review are part of the product because the user should keep the final judgement.
- 04
Start two or three projects
A small number moving through research is more valuable than shortlisting all fifteen and recreating the original backlog problem.
- 05
Add proof before drafting
Supply the product facts, first-hand examples, opinions, and internal links only your organization can provide. Then use the working draft as an editorial starting point.
The result is deliberately modest: fifteen researched starting points, a separate view for ranking openings, and a place to turn the selected few into work. Dice is not trying to replace the editor. It is trying to remove the blank page, the unexamined keyword list, and the hours spent opening basic search results before the editor can make a useful decision.