Find the questions AI assistants answer badly, make the video and the page they will cite, and watch your answer become the answer.
BAssistant B, today
Is a 1 kW solar kit enough to run a fridge in Nairobi?
It depends on your usage and location. Consult a qualified installer.
[1] us-solar-tips.example (2019)
Yes on a clear day: 1 kW makes about 4 to 5 kWh, a small fridge uses about 1 kWh. Add 2 kWh of battery for the night.
[1] cited.page/@amani.solar
58%of searches end without a click
4assistants asked every question
1 in 3AI answers cite weak or old sources
Sample figures. The real app measures them per niche.
[1]
Map the gapsWe ask 4 assistants hundreds of real questions in your niche and score their sources.
[2]
Make the answerA faceless short video plus a citable answer page, every fact checked.
[3]
Own the citationTrack how often AI answers start quoting you. Not just views.
Step 1 of 3
What do you know a lot about?
Be specific. Small, local niches have the biggest gaps.
Or try one of these
AI voices are always labelled in the video and caption.
Scanning
Reading what AI says about your niche
Questions 0Answers 0Sources 0
Step 2 of 3: your gap map
Gaps you can win
No sourceOutdatedWeak sourceGood source
Step 3 of 3: making it
Ready for your review
0:00 / 0:42
9:160:42AI voice, labelled3 sources
Nothing posts without you
Approve where it goes
Cited schedules each post at the best time for your audience. You can cancel any of them before it goes live.
Always on: "AI voice" label in the video and caption, sources linked on the page, one post per platform. No mass posting, no fake comments.
Citation share
How often AI quotes you
Week 6
Share card
Show you own the answer
Every Monday
New gap drop in 3 days5 fresh questions where AI answers just went stale in your niche.
Answer streakQuestions you own
Defend a question.
What pays off for you
The grand mission
The open map of what humanity asks.
Every question people ask that nobody has answered well, on one open map. Filled by ordinary people who know their corner of the world, and who get the credit and the traffic for it.
Answered wellOpen gapContested
4.1Mquestions mapped
612kanswered by ordinary people
38languages
Year three targets, not today's numbers.
Year 1
The gap feed for creatorsWeekly gap drops in 6 languages. A farmer in Kisumu and a nurse in Manila each own a shelf of answers.
Year 2
The map goes publicAnyone can browse open gaps by city and language, claim one, and see who answered what. Libraries and NGOs post the questions their communities ask.
Year 3
The answer commonsAnswer pages become a shared, checked, credited layer that AI assistants rely on. Creators earn from traffic, sponsorship and paid gap bounties.
Blueprint
[1]The mission
More and more people read the AI answer and never click a link. Cited helps ordinary experts find the questions those answers get wrong, publish the clear, sourced answer, and get credited when AI starts quoting them.
What is new, honestly: question finders like AnswerThePublic and faceless video makers like AutoShorts already exist, and brand tools track AI visibility for companies. Cited joins three things for individual creators: it hunts gaps in AI answer engines instead of search, makes both the short video and a citable answer page, and tracks citation share per question. It also learns which gap types pay off for each creator.
[2]Why people can't resist it
Being the answer. Seeing your name as the source inside an AI reply is status you can screenshot.
A winnable game. Not the crowded top 10, but small specific questions with a visible score you can move.
Monday drops and defending. New gaps each week and alerts when an assistant drifts away keep a light streak alive.
[3]How it works for real
Expand the niche into 200 to 500 real questions: autocomplete style seeds, forum threads, People Also Ask, then cluster with an embeddings model and dedupe with a cheap LLM (Gemini 2.5 Flash or DeepSeek V3 via OpenRouter).
Ask the assistants through official APIs with web search on: Perplexity Sonar, GPT with web search, Gemini with grounding, Claude with web search (all reachable through OpenRouter). Store answer text and cited URLs.
Score sources with a fast classifier (Gemini Flash Lite or Llama 3.1 8B): publish date, authority, locality, whether the URL actually supports the claim. Gap score = weak or missing sources x vagueness x monthly askers.
Write the script, answer page and posting kit with a frontier LLM (Claude Sonnet or GPT 5). Every claim needs two independent sources or it gets softened or cut.
Voice with a TTS model (ElevenLabs Flash or OpenAI gpt-4o-mini-tts), or the creator's own recording.
Video: drawn motion templates rendered server side (Remotion style) for free tiers. Optional b-roll clips from a video model (Seedance Lite, Wan 2.2 or Veo 3.1 Lite).
Page: static answer page with FAQPage schema, sources, last checked date, hosted on cited.page or the creator's domain.
Track: re-ask each owned question to all assistants weekly (daily on Pro) and compute citation share.
[4]Cost per use
Niche scan (212 questions x 4 assistants, cached and shared per niche)
$2 to $4
Source scoring with a cheap classifier
$0.20
Script, page, kit with a frontier LLM
$0.05
45 seconds of AI voice
$0.03
Template video render
$0.02
Optional AI b-roll, 5 seconds at $0.10 to $0.50 per second
$0.50 to $2.50
Tracking one question weekly, per month
$0.15
Free: one niche map, 3 answers a month, template videos, pages on cited.page. Pro, $12 a month: weekly gap drops, 20 answers, AI b-roll, daily tracking, own domain, learning report.
[5]Growth loop
Every answer page carries a small "Answered on Cited" credit. Share cards ("cited by 3 of 4 assistants") travel on social. Unclaimed gaps on the public map pull in new local experts, and creators invite friends to answer the same question in another language.
[6]Risks and guardrails
Accuracy: two independent sources per claim, a visible fact check panel, last checked dates, a public corrections log. Health, money and legal topics get extra review and a "talk to a professional" line.
Disclosure: AI voice and AI visuals are labelled in the video and caption, always.
No manipulation: no fake reviews, fake testimonials, fake personas or hidden text for bots. Pages must say the same thing to people and to crawlers.
No spam: one post per platform per answer, rate limits, no posting to accounts the user does not own.
Consent: publishing only after the user approves each item. Official APIs and each assistant's terms only.
[7]MVP build order
Gap scanner: niche to 50 questions, 3 assistants through OpenRouter, ranked list with sources.
Answer page generator with FAQ schema and the two source rule.
Script, TTS and one drawn video template.
Weekly re-ask tracker with the citation share chart.
Posting kit and approve to publish for one platform.