The Read Option AI Intelligence Manifesto and Trust Rules
The Read Option AI Intelligence Manifesto
Read Option AI exists to help business owners make better decisions.
Every recommendation, report, score and integration in Read Option AI must satisfy these ten principles. They guide every feature we build next.
- 01
Evidence before opinion.
Nothing is asserted before a source has been read. The receipts come first and the conclusion follows them.
- 02
Transparency before confidence.
Read Option AI shows how it reached an answer before it tells you how sure it is. You can inspect the work on every screen.
- 03
Observed before estimated.
A measured figure always outranks a modelled one, and an estimate is always labeled as an estimate.
- 04
Explain every recommendation.
Every play, score and number carries a plain-English explanation of where it came from.
- 05
Never fabricate business intelligence.
Promotions, campaigns, social activity, reviews, website changes, financials and engagement metrics are reported only when observed.
- 06
Unknown is better than incorrect.
When the evidence is not there, Read Option AI withholds the claim and says so rather than filling the gap.
- 07
Confidence increases with evidence.
Confidence is earned by corroboration and coverage — thin sourcing lowers it automatically.
- 08
Recommendations should always improve as Read Option AI learns more.
Every connection, scan and result sharpens the next recommendation, and Read Option AI tells you what would sharpen it further.
- 09
Every recommendation answers four questions.
What? Why? How do we know? What should I do next?
- 10
Read Option AI is not an AI that guesses. Read Option AI is an AI that scouts.
Scouting is observation with sources and dates. Guessing is not a product feature.
Every recommendation answers
What?
The recommendation itself, stated plainly.
Why?
The business reason it matters right now.
How do we know?
The source, the observation, the date and the confidence.
What should I do next?
The concrete next action the owner can take today.
Read Option AI is not an AI that guesses. Read Option AI is an AI that scouts.
- 01
Never fabricate data.
A number is shown only when a registered source actually reported it. Nothing is filled in to complete a screen.
- 02
Never estimate a metric without clearly labeling it as an estimate.
Calculated figures carry the “Estimate” label and disclose their inputs and assumptions.
- 03
Never display competitor data unless it was actually retrieved.
Competitor figures appear only when Read Option AI read them from a public listing or page it can cite.
- 04
Never infer revenue from website traffic.
Revenue comes from connected financial data. Traffic is reported as traffic and nothing more.
- 05
Never infer customer satisfaction from review count alone.
Review volume is reported as volume. Satisfaction requires ratings or review content.
- 06
Always identify the originating data source.
Every metric carries a source badge naming where the value came from.
- 07
Always show when data was last synchronised.
Each reading shows its sync time, or states that no sync time is recorded.
- 08
If data is unavailable, say so explicitly.
The surface shows “Read Option AI does not currently have enough verified information.” and what would be needed to answer.
- 09
Protect trust above all else.
Withholding an answer is always preferred to producing a confident-sounding guess.
When data is unavailable, Read Option AI says exactly this: “Read Option AI does not currently have enough verified information.”
Read Option AI Trust Rules
Trust and transparency are more important than providing an answer.
These rules apply to every score, recommendation, report and AI answer in the platform. If following them means Read Option AI cannot answer, Read Option AI does not answer.
The three intelligence types
Every source and every recommendation is labeled with exactly one type. They are never mixed together.
Connected Intelligence
Data the business owner authorized Read Option AI to access.
Read from accounts and answers you authorised. These are your own numbers, not an approximation of them.
Public Intelligence
Data collected from publicly available sources.
Observed from what anyone can see online — your website, listings, reviews and public profiles. Nothing here required an account link.
Estimated Intelligence
Industry benchmark calculations.
Calculated from typical figures for businesses like yours because the real number could not be measured. Treat it as a range, never as fact.
The four intelligence levels
Every recommendation displays which level of intelligence produced it.
Level 1 · Public
Read from what anyone can see about the business online.
Google Business Profile · Website · Reviews · Scouting Report · PageSpeed · Social Media
Level 2 · Connected
Read from accounts the owner authorised Read Option AI to access.
Google Business Profile · Google Analytics · Search Console · Meta · QuickBooks
Level 3 · Operational
Read from the systems that run the business day to day.
Accounting · Scheduling · POS · Job Management · Inventory
Level 4 · Executive
Analysis calculated from verified evidence using published rules.
Play prioritization · Evidence-based impact explanations
Read Option AI must never invent
Promotions
Never invent a promotion, offer, discount or deal.
Only promotions quoted from a page Read Option AI actually read are shown.
Campaigns
Never invent a campaign a business or competitor is running.
Campaigns are reported only when a public artifact evidences them.
Social media activity
Never invent posts, cadence, followers or social presence.
A platform that was not located reads “Public profile not found.”
Review activity
Never invent reviews, ratings, review counts or review trends.
Review figures come from the listing or API that reported them.
Website changes
Never invent website changes, redesigns or new pages.
A change is only claimed when two reads of the site differ.
Financial information
Never invent revenue, spend, pricing or financial performance.
Revenue figures come from connected billing data, or are labeled as estimates.
Engagement metrics
Never invent impressions, clicks, opens, reach or engagement rates.
Metrics come from a connected account, or they are withheld.
When evidence is unavailable
Read Option AI says so. It withholds the claim and shows “Not enough publicly available evidence.” together with what it would need in order to know.
When confidence is low
Read Option AI explains why — thin coverage, a single source, an inference from benchmarks rather than a direct observation.
When public information cannot be verified
Read Option AI displays “Not enough publicly available evidence.” rather than characterising the business at all.
When public information cannot be verified, Read Option AI displays exactly: “Not enough publicly available evidence.” — no placeholder numbers, no illustrative examples, no filler.
AI safety rules
AI never decides what is true in Read Option AI. It explains conclusions the evidence already reached — and it is forbidden from filling any gap it finds.
AI must never: Estimate missing data.
A missing reading stays missing and is named as missing.
AI must never: Guess customer metrics.
Only figures supplied by a connected, validated source are used.
AI must never: Invent trends.
A trend requires two comparable periods of verified data.
AI must never: Create fictional competitors.
Competitors appear only when a public listing was actually read.
AI must never: Fabricate marketing performance.
Impressions, clicks, opens and spend come from the platform that reported them.
AI must never: Generate unsupported recommendations.
The recommendation is withheld and Read Option AI says: "We don't have enough information about your business to recommend this yet."
When the data is unavailable, Read Option AI says exactly: “We don't have enough information about your business to recommend this yet.” Read Option AI will always prefer “I don't know yet.” over a guess.
The mandatory workflow
Before any recommendation is shown to you, anywhere in Read Option AI, all nine steps below must complete in code. If a step cannot complete, the recommendation is withheld instead of generated.
- 1Collect data from every connected integration
- 2Validate the data
- 3Identify supporting evidence
- 4Cross-check evidence across integrations
- 5Calculate confidence
- 6Estimate business impact
- 7Estimate implementation effort
- 8Rank against every other recommendation
- 9Explain the recommendation
AI is never allowed to determine a recommendation. It may only explain a decision the Read Option AI Evidence Engine has already made.