From a keyword to a live article, on autopilot.
Connect a domain and Vupie analyzes the business, the market and your voice in about twenty minutes. After that, every article walks the same ten steps: planning from Search Console, research from primary sources, review on a model from a different vendor, a score against that source set, publication to your CMS, and index verification. Each step writes to the article record, and the record is visible in your dashboard.
Seven-day trial when your spot opens. One-click cancel. No setup fee on any plan.
The pipeline
Ten steps between a topic
and a published article.
Every article walks the same ten steps. Each step writes what it did to the article record, and you can open any article in the dashboard to read that record.
Planning from Search Console
Planning runs on your own data: queries you already get impressions for, positions 8 to 30, and questions that draw impressions with no page that answers them. Results-page data is a direct lookup, so it is used to check who ranks now and what the current top ten leave unanswered. Volume estimates are shown as low, medium or high, because in a study of 72,635 keywords volume was overestimated in more than 90 percent of cases.
Research from primary sources
Candidate sources are retrieved and classified before drafting, and only material that passes classification enters the source set the model writes from. The classes that count, and the ones that do not, are listed under sourcing below. The citation list travels with the article.
Outline
Headings, the question each section answers, and the internal links to place. The outline is also where the synthesis requirement described under sourcing is enforced.
Drafting in your voice
Written section by section from the retrieved sources, against your voice profile and your experience bank. The model may cite only from that source list. Each section opens with an answer that stands on its own.
Cross-vendor review
Review runs on a model from a different vendor than the one that wrote the draft. Cross-family verification catches errors a model does not catch in its own work, and the benchmark behind that choice is in the gate section below. Claims the sources do not support are rewritten or removed.
Internal and network linking
Links to your existing pages are placed automatically. A network placement is added when a topically matching paragraph and an eligible receiving site both exist. When there is no relevant match, the slot stays empty and an external source takes its place. Every placement is written to your ledger.
Images and assets
Hero image, in-article visuals, alt text and structured data are generated and attached. AI-generated images are written out with IPTC DigitalSourceType metadata, so the provenance travels with the file.
Quality gate
The draft is scored from 0 to 100 against the review dimensions listed in the gate section below. Below threshold, the weak sections are rewritten, up to twice. An article that still fails is refused.
Publish and notify
Published to your CMS at the time you choose. The sitemap is pinged and IndexNow is notified, so the new URL is announced rather than waited on.
Index verification and monitoring
Indexing is checked through Google's URL Inspection API on days 2, 7 and 14. Every article carries a status: indexed, or not indexed with a reason and a remedy. Clicks and impressions are then tracked in Search Console, and pages that start losing ground are picked up by the refresh engine.
Capacity is eight articles a month on Starter, twenty-four on Growth, and sixty pooled across ten client sites on Agency. The delivered number is whatever passes the gate, and every refusal is visible with its reason.
Information sourcing
Where every fact
comes from
Structure is not quality. A spam article and a good one both have a summary box, clean headings, a comparison table, five citations, and 1,400 words. Generative Engine Optimization research shows that visibility gains come from quotations (+41%), statistics (+30%), and citing sources (+27%).
Vupie sources these in four places.
01
Your own numbers
We compute publishable figures from what is already connected: catalog and order data, Analytics, Search Console, your review corpus. Price distributions, seasonality, recurring themes in customer feedback. You approve each figure before it can be used, and the method travels with it into the article. These are statistics no competitor can publish.
02
Your own judgment
For each article, the system writes two or three specific questions the piece needs a practitioner to answer. They arrive by email, SMS, or Slack, and take under two minutes to answer from a phone, including as a voice note. Answers enter the article under your byline and are retained, so they compound across future articles on adjacent topics.
03
Primary sources
Research classifies every candidate source. Standards, statute, regulatory filings, peer-reviewed work, official statistics, vendor documentation, and live pricing pages count. Trade press with a named author counts. Listicles and content written to rank do not count and never enter the source set the model writes from. Minimum two primary sources per article.
04
Synthesis nobody has attempted
Sometimes the contribution is the comparison itself. Every outline must name what the article says that the current top ten do not, and that claim is checked against the results page rather than taken on the model's word.
Review
The gate
Review runs on a model family different from the writer, for example Claude auditing Gemini. Cross-family verification yields 52.15% accuracy on hard factual benchmarks compared to 44.74% for the best single model, a gain of 7.41 percentage points at p<0.0001.
Review scores factual coverage against sources, citation coverage, originality against ranking results, brand voice, and substance density measured as verifiable claims per 100 words. Below threshold, weak sections are rewritten up to twice.
Refused articles appear in your calendar with the reason in plain language. The topic returns to planning and is not selected again until input conditions change.
Length is an output, not a target. Nine hundred words carrying fourteen real claims beats 2,200 words carrying the same fourteen.
Five conditions fail an article regardless of its overall score:
- 1 No demonstrable contribution beyond what currently ranks
- 2 Fewer than two primary sources
- 3 No original data, answered expert question, or first-hand material
- 4 Any claim contradicted by its own source
- 5 Substance density remaining below threshold after an edit pass
Video
Every article can carry
an explainer video.
Four formats, chosen per topic, running 45 to 60 seconds. The video is embedded in the article, marked up with VideoObject schema, and delivered as an MP4 that exports directly to Shorts, Reels, TikTok and LinkedIn.
Narrated slides
The article's key points as branded slides with a synthetic voiceover. The cheapest format to produce, and it fits most topics.
B-roll cut to narration
Licensed footage edited against the script, for topics where showing something is clearer than describing it.
Avatar presenter
A synthetic presenter reads the approved script to camera. Labelled as synthetic on the page and in the file metadata. This format is a disclosed add-on.
Subtitles and transcript
Subtitles in the article's language and the full transcript on the page, so the material is readable and searchable without sound.
What the video is for
- A second asset from one research pass, produced from the approved script
- A file you own and can publish on your own channels, including YouTube
- A transcript on the page, so the material is readable and searchable without sound
- VideoObject schema, which describes the video to search engines
What we do not claim
We do not claim embedded video improves web search rankings. Google shows video thumbnails only when video is the main content of the page. We produce video because it functions as an on-site conversion asset and because you can use it on your own channels.
Video credits are included on the Growth plan.
Brand voice
A voice profile built from your pages and your edits.
The profile is built from your published pages. Answers to the practitioner questions described under sourcing feed it too: field tests, customer cases and opinions enter the experience bank and are reused across articles on adjacent topics. Every edit you make to a draft updates the profile.
- Versioned voice profile, with a diff between versions and a revert
- Experience bank: products, customer cases and answered questions, reused across articles
- Every statistic cited to a named source
- Drafts below threshold are rewritten or refused, and the reason is recorded
Voice profile, acme-consulting.example
v7 activeVoice & tone
v7 · fewer rhetorical questions in intros
from your editsv6 · name the assumption behind every number
from your editsExperience bank
Quality gate
What a discovery phase actually costs
92/100 · publishedConsultant day rates in 2026
64/100 · heldAfter two failed revision passes the article is refused. The reason appears in your calendar and the topic returns to planning.
Provenance
Every article carries a record of how it was made.
Marking, sources and disclosure are part of the output.
AI images carry provenance metadata
Every generated image is written out with IPTC DigitalSourceType metadata. It is machine readable, so anyone inspecting the file can see how the image was produced.
Sources are published with the article
Each article ships with the list of sources it cites, linked. Any statistic can be followed back to the source it came from.
Disclosure is set per site
Choose how articles disclose AI involvement: a line in the byline, a note in the footer, or the machine-readable marking alone. The setting is recorded with the article.
Each article keeps a signed record of which models were involved, when, from which sources, and who approved it. One named person holds editorial responsibility per site. If the system modifies an article after approval, the approval lapses and has to be given again. Data is hosted in the EU and a data processing agreement is available at signup. We do not claim this makes you compliant with the EU AI Act, and the limits section below spells out why.
AI visibility
See where ChatGPT, Perplexity
and Gemini cite you.
Run the same prompt through ChatGPT three times and roughly 2.2 to 2.3 percent of the cited sources are stable across runs. A 2026 variance study of that class of answer found that brand identity accounts for 1.5 percent of the variation, the language of the query for 26.5 to 32.0 percent, and repeating the same phrasing for 35 percent. A single query check provides no reliable signal, and that is what a daily single-prompt check measures.
So the measurement budget goes to languages and phrasings instead of repeats. One additional language reduces measurement error roughly fifteen times more than one additional repeat. Design depth scales with your plan.
The traffic is worth the trouble. Across retail traffic in Q1 2026, AI-referred visitors converted 42 percent higher and generated 37 percent more revenue per visit compared to standard search traffic. Tracking those visits requires distinct measurement parameters.
Monthly measurement
acme-consulting.exampleVisibility
24%CI 18 to 31
What moves the answer
Monthly measurement with a confidence interval
The full measurement runs monthly across engines, languages and phrasings, and is reported with a confidence interval. A weekly pulse runs between them and is labelled as directional. No other tool in the category publishes its uncertainty.
Share of voice per engine
For every tracked question we record which brands the engines recommend and which pages they cite, per engine, over time, next to the competitors you name.
AI referrals in GA4
A GA4 segment isolates sessions arriving from ChatGPT, Perplexity, Gemini and other assistants, so referral traffic is counted rather than estimated.
AI-crawler access audit
We fetch your key pages as GPTBot, ChatGPT-User, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended and check robots.txt and firewall rules. Bot-management defaults block these agents on many sites with no visible sign. The audit reports the block and the fix for your setup, and it also runs before the first article is published.
Draft-time citability score
Every draft is scored before it ships on the levers with experimental support: direct quotations, statistics with attribution, inline source citations, and sections whose opening sentences stand alone as an answer. Schema markup is used for search features, not as a citation lever.
Weekly decay scan
Auto-refresh onRanked by clicks lost.
Refresh and decay
Weekly decay scanning, with automatic refresh.
AI engines cite by last-updated date. Across 7,683 pages and 47,097 citations, Gemini cited content updated within the year 78 percent of the time, ChatGPT 73 percent and Perplexity 65 percent, while only 42 percent of cited content had been recently published. Vupie scans your published articles in Search Console each week, finds the pages losing ground, re-runs the research, updates figures and sources, republishes, and verifies the index again.
- Weekly scan of every published article
- Pages ranked by the clicks they are losing
- Research re-run, figures and sources updated, article republished
- Sitemap pinged and IndexNow notified on republish, index checked again on days 2, 7 and 14
Publishing
Thirteen publishing destinations.
Native APIs where they exist, a webhook or REST API where you would rather wire it yourself, and a hosted blog if you have no CMS.
WordPress
Publishes via the REST API, categories, tags, authors and featured image set for you.
Shopify
Posts to your store blog with your theme's styling and product links intact.
Webflow
Writes items to your CMS collection and triggers a site publish.
Wix
Creates posts through the Wix Blog API, cover image included.
Ghost
Publishes through the Admin API with tags and feature image handled.
Framer
Adds entries to your Framer CMS collection, ready to go live.
GoHighLevel
Posts to the blog on whichever sub-account you choose.
Duda
Publishes to your site's blog through the official Duda API.
BigCommerce
Store-blog publishing with meta fields and URLs handled.
Snapps
Native publishing straight to your Snapps site's blog.
Webhook
We POST the full HTML and metadata to any URL you give us.
REST API
Pull finished articles into your own stack, on your own schedule.
Hosted blog
For sites with no CMS. Vupie hosts the blog on your own domain, see below.
Hosted blog on your own domain
Point blog.yourdomain.com at us with one CNAME record and the whole blog runs on Cloudflare's edge, sitemap and robots.txt handled. If you cancel, it keeps serving for twelve months and exports whenever you want it.
Limits
What we don't claim
That video improves search rankings.
No dwell-time signal is documented in Search schemas. Video is provided for conversion and off-site distribution.
That Google penalizes AI content.
A study of 1,000,000 pages across 100,000 results pages found that top-10 positions contain 8.4% to 11.7% pages that are at least 80% AI generated, with stable impressions over twelve months. Google acts on templated narratives from shared infrastructure at scale (S-CTS). Vupie varies templates, publishing frequency, and link structures per customer.
That fewer articles rank better.
No study proves this. We write fewer articles because quotations, statistics, and primary citations are what generative engines cite, and retrieving them requires work.
That we make you compliant with the EU AI Act.
Article 50(4) applies to text published to inform the public on matters of public interest where no human holds editorial responsibility. Commission guidelines state that routine marketing content falls outside scope on the public interest provision. We provide signed provenance records, named editorial responsibility, EU hosting, and a Data Processing Agreement.
Traffic within a fixed number of days.
We report position metrics in Search Console and signal changes per phase.
What happens when
you connect a site.
Enter your domain. Vupie analyzes the business and returns a plan in about twenty minutes. The first articles arrive with their sources and their scores before you commit to anything. Vupie is in closed beta, so registration joins the waitlist and we onboard in waves.
Get startedSeven-day trial when your spot opens. One-click cancel. No setup fee on any plan.