The platform field guide.

A personalized feed, not a single engagement leaderboard. Read this entry as a historical map of disclosed stages, not today's secret formula.

Disclosure boundary. Historical Meta system card (2022). Feed candidates, predictions and ranking are described at a high level; current weights and behavior have not been verified. This is not Reels or Explore guidance.

Read the disclosed signals & their limits

Meta's historical Feed card describes gathering eligible content, using signals to predict likely interactions, and ranking candidates for a particular viewer.

Context. Instagram Feed in the historical 2022 disclosure.

What this does not establish

Not a current 2026 specification or a complete candidate-selection policy.
Do not apply this Feed description to Reels or Explore.

Meta (2022)

The historical disclosure describes information about content and a viewer's past activity and relationships as inputs to predictions, rather than a universal score visible to creators.

Context. High-level signal families in the historical Feed system card.

What this does not establish

No exact signal weights or causal return from obtaining a like, save or comment are established here.
Creator-visible engagement is not a measurement of the internal ranking model.

Meta (2022)

Your choices / Not algorithm instructions

What to make

Make the first image and caption identify who the post is useful for and what it delivers.
Keep the visual, caption and landing promise consistent; use readable text and meaningful image descriptions.
Choose one useful action such as saving an actual reference, not an engagement-bait demand to comment.

An outcome to define / Not a score

Save rate (unique-account basis)

Count this outcome: Reached accounts that saved this post, counted once in the observation window
Out of this eligible group: Accounts reached by the same post in the same window

Use this only if the same unique-account unit is available for both counts. Aggregate repeat events are not unique savers; unavailable counts stay blank rather than being inferred. Reach is not randomized exposure.

Quality & trust

What must not get worse

The saved post contains the promised reference or useful information.
Review hides, unfollows and feedback about misleading expectations when available.
Do not reward an engagement rate that comes with inaccessible text or a bait-and-switch.

Confounders

What else might explain it

Follower and non-follower mix
Relationship history and personalized exposure
Posting time, competing content and paid distribution
Format changes or unequal reach between posts

Source dates, access & references

Meta
Instagram Feed ranking — historical system card (opens in a new tab)
Published: 2022 · Editorial review: 2026-09-08
Historical 2022 disclosure of Instagram Feed candidate selection, predictions and ranking. The scope is Feed, not Reels, Explore or all of Instagram.
Access & limits. Historical reference, not a freshly verified 2026 specification. The URL may now present a changed document. No current signal weights, complete model, or outcome guarantee are established by this record.

02 /

Platform × surface

TikTok For You

For You

A recommendation surface distinct from Search. The available reference supports only a conservative, high-level account of its signals.

Disclosure boundary. Official index only: the linked article body was access-restricted during the approved research. Current exact wording, weights and a complete For You pipeline were not verified.

Read the disclosed signals & their limits

TikTok's official help-index information groups recommendation inputs into user interactions, content information and user information.

Context. Official index-level information; direct article-body access was restricted.

What this does not establish

Index only, not a freshly verified reading of the full For You documentation.
No exact weights, watch-time threshold or guaranteed distribution sequence is known from this record.
A broad signal family does not establish which creative tactic causes exposure.

TikTok Help Center

Making a video's topic and payoff clear early is a sensible creative hypothesis to test against actual viewing behavior, not an official ranking command.

Context. A lab planning choice under limited source access.

What this does not establish

The source does not validate this template or a particular opening length.
Recommendation exposure changes the audience seeing each variant.

TikTok Help Center

Your choices / Not algorithm instructions

What to make

Use one legible opening that names the activity or question; compare it with a direct answer.
Keep the promised demonstration in the video rather than delaying it until a separate post.
Use captions and a coherent visual sequence; keep pace and duration fixed when testing the opening.

An outcome to define / Not a score

Full-watch rate

Count this outcome: Eligible video starts that reached the end, at most once per counted start
Out of this eligible group: Eligible starts of the same video in the same reporting window

Use one analytics definition for both variants. Replays, loops and platform-defined views may differ from starts; do not mix their counts. Completion is not satisfaction or proof of recommendation causality.

Quality & trust

What must not get worse

The ending resolves the opening promise.
Check substantive feedback and confusion, not only completion.
Avoid strobing, artificial distress and forced rewatch loops.

Confounders

What else might explain it

Personalized audience selection and distribution
Topic interest, sound or trend changes
Duration, pacing and replay behavior
Posting time, moderation and competition

Source dates, access & references

TikTok Help Center
How TikTok recommends content (opens in a new tab)
Published: date not established · Editorial review: 2026-09-08
Official help-index information about recommendations, including For You and Search. These are separate surfaces, not one interchangeable algorithm.
Access & limits. Official index only: direct access to the article body was restricted during the approved research. Broad signal families are recorded conservatively; exact wording, current weights and a complete surface-specific pipeline were not verified. No new retrieval was performed on the editorial review date.

03 /

Platform × surface

TikTok Search

Start with an explicit query and the answer a searcher needs. Do not treat a For You hook as a documented Search-ranking factor.

Disclosure boundary. Official index only: the help article body was access-restricted. Search is a separate surface; no verified full retrieval pipeline, exact Search weights or keyword-volume data are supplied here.

Read the disclosed signals & their limits

TikTok's official recommendation help index includes Search as a distinct surface. That does not make its ranking interchangeable with For You.

Context. Official index-level surface information, with direct body access restricted.

What this does not establish

Index only; detailed, current Search-specific weighting was not verified.
This entry cannot tell you query volume, likely rank or a guaranteed discovery method.

TikTok Help Center

Matching the visible question, on-screen explanation and demonstrated answer to one search intent is a planning hypothesis, not a known formula for ranking.

Context. A search-oriented content brief proposed by this lab.

What this does not establish

The limited reference does not prove a causal effect of a keyword in any specific field.
Do not claim search demand or ranking gains without your own appropriately scoped observations.

TikTok Help Center

Your choices / Not algorithm instructions

What to make

Write the query as a real user question, then show the answer in the same video.
Keep the spoken explanation, caption and on-screen topic accurate and consistent, without keyword stuffing.
Record query, region, date and search context when manually observing results; do not assume another user's results match.

An outcome to define / Not a score

Search-result selection rate (when measurable)

Count this outcome: Eligible Search-result presentations that led to viewing this result
Out of this eligible group: Eligible presentations of this result for the same query cohort and window

Use only if a consistent search-specific exposure denominator is actually available, or measure it in a controlled search-task study. A share of views attributed to Search is not a click-through rate. Missing exposure data means this rate is unavailable.

Quality & trust

What must not get worse

A viewer can answer the original query after watching.
Do not use an irrelevant trending term to acquire mismatched visits.
Inspect query-specific confusion and whether the answer remains current.

Confounders

What else might explain it

Query mix, region, language and personalization
Unknown result exposure and rank changes
Search demand, competing results and trend shifts
Differences in analytics attribution and observation window

Source dates, access & references

TikTok Help Center
How TikTok recommends content (opens in a new tab)
Published: date not established · Editorial review: 2026-09-08
Official help-index information about recommendations, including For You and Search. These are separate surfaces, not one interchangeable algorithm.
Access & limits. Official index only: direct access to the article body was restricted during the approved research. Broad signal families are recorded conservatively; exact wording, current weights and a complete surface-specific pipeline were not verified. No new retrieval was performed on the editorial review date.

04 /

Platform × surface

YouTube Home

Home

Recommendations respond to a viewer's history and satisfaction-related information. Make the title-thumbnail promise match the video rather than optimizing clicks alone.

Disclosure boundary. Official high-level recommendation guidance, not a complete algorithm. Home relies primarily on watch history; disclosed signals vary by surface and no fixed weights are supplied.

Read the disclosed signals & their limits

YouTube says Home recommendations rely primarily on watch history, within a broader recommendation system using signals such as viewing activity, subscriptions and feedback.

Context. YouTube Home, not the Shorts feed or a search-ranking specification.

What this does not establish

Different recommendation surfaces use signals differently.
The document does not reveal a full candidate pipeline or exact signal weights.

YouTube Help

YouTube's recommendation explanation includes satisfaction surveys and user feedback; watch time alone is not a full account of viewer value.

Context. Official explanation of recommendation inputs, not a creator-visible score.

What this does not establish

Creators cannot reconstruct internal satisfaction predictions from their click rate.
A single video's analytics do not identify why the recommendation system changed exposure.

YouTube Help

Your choices / Not algorithm instructions

What to make

Make one specific title promise and use a thumbnail that accurately represents the video.
Deliver evidence or a demonstration early, with a clear route through the rest of the video.
When testing a title frame, hold the thumbnail and video fixed rather than silently testing a different package.

An outcome to define / Not a score

Home impression click-through rate

Count this outcome: Views attributed to eligible Home thumbnail impressions
Out of this eligible group: Eligible Home thumbnail impressions in the same window

Use Home-filtered data when available; broader Browse features are not necessarily Home alone. Total views, external views and all-surface impressions are not interchangeable. CTR can change when YouTube broadens the audience.

Quality & trust

What must not get worse

Review retention and whether the promised answer actually arrives.
Inspect feedback about misleading titles, including disappointment despite a click.
Do not sacrifice accuracy or accessibility to a more selectable thumbnail.

Confounders

What else might explain it

Viewer watch history and recommendation exposure
Audience breadth, subscriber mix and returning viewers
Topic demand, competing videos and seasonality
Thumbnail, title or video changes made together

Source dates, access & references

YouTube Help
How YouTube recommendations work (opens in a new tab)
Published: date not established · Editorial review: 2026-09-08
Official overview of recommendation signals and differences between surfaces. Home relies primarily on watch history; recommendations also use feedback and satisfaction information rather than watch time alone.
Access & limits. Living help document; publication date and exact weights are unknown here. Reviewed against the approved reference, not freshly fetched. Home guidance must not be substituted for Shorts or search ranking.

05 /

Platform × surface

YouTube Shorts

Shorts feed

In a swipeable feed, choosing to watch, continued viewing and satisfaction are different questions. Test an opening without changing the lesson it leads to.

Disclosure boundary. Official Shorts discovery guidance describes personalization and performance signals, including choosing to view, retention and satisfaction-related feedback. It does not publish a universal weighting formula.

Read the disclosed signals & their limits

YouTube's Shorts guidance describes whether viewers choose to watch, average view duration and average percentage viewed as signals considered in discovery.

Context. Shorts recommendation and discovery guidance.

What this does not establish

No universal duration, completion threshold or guaranteed distribution is disclosed here.
These are distinct signals, not interchangeable denominators for one viral score.

YouTube Help

The guidance also discusses likes, dislikes and post-watch survey information, while noting topic interest, competition and seasonality as influences on reach.

Context. Shorts satisfaction-related feedback and discovery context.

What this does not establish

High performance on one measure does not guarantee impressions.
Survey inputs and ranking predictions are not fully visible in creator analytics.

YouTube Help

Your choices / Not algorithm instructions

What to make

For a tutorial, compare a direct-answer opening with one concrete question; use the same demonstration after the opening.
Keep framing, captions, duration, audio and CTA fixed so the opening is the treatment.
Resolve the question before asking for any optional next action; do not hide the answer in a forced loop.

An outcome to define / Not a score

Chose-to-view rate

Count this outcome: Eligible Shorts-feed presentations recorded as choosing to view rather than swiping away
Out of this eligible group: Eligible Shorts-feed presentations with a recorded view-or-swipe decision in the same window

Use the same viewed-versus-swiped-away definition for both variants. Total views, replays and feed presentations are different units; do not divide all-platform views by feed exposures. Verify the analytics definition before entering counts.

Quality & trust

What must not get worse

Check retention or completion with a consistent definition and unchanged duration.
Read feedback about whether the actual lesson satisfied the opening promise.
Reject apparent gains caused by confusing loops, withheld answers or inaccessible presentation.

Confounders

What else might explain it

Topic interest, competition and seasonality
Audience mix selected by recommendations
Duration, repeat views and changing view definitions
Posting time and unequal distribution

Source dates, access & references

YouTube Help
Search and discovery tips — Shorts (opens in a new tab)
Published: date not established · Editorial review: 2026-09-08
Official explanation of Shorts discovery, viewer personalization, choosing to view, retention and satisfaction-related signals; topic interest, competition and seasonality also affect exposure.
Access & limits. Living help document; publication date and model weights are not supplied here. Reviewed against the approved research reference, not freshly fetched. The disclosure is directional guidance, not a deterministic reach formula.

06 /

Platform × surface

Google Search

Web Search

Build the page that answers the query with original, reliable evidence. Content guidance and canonicalization are useful, but neither is a ranking guarantee.

Disclosure boundary. Official people-first guidance is not a complete retrieval or ranking specification. Canonicalization concerns choosing representative URLs among duplicates; it is not an emotional-engagement signal.

Read the disclosed signals & their limits

Google recommends helpful, reliable content made for people, including original information, clear sourcing and a satisfying experience for the intended reader.

Context. Search Central content-quality guidance, not a list of independent ranking weights.

What this does not establish

The guidance does not establish a fixed ranking formula or guarantee indexation and rank.
E-E-A-T is not a single ranking factor; a checklist score here would be invented.
This registry provides no live query volume or competitor measurements.

Google Search Central

Google may select a representative canonical URL among duplicate or very similar pages; the publisher's declared canonical is a signal rather than an instruction that must be followed.

Context. URL consolidation and indexing, separate from the creative appeal of a title.

What this does not establish

A canonical declaration does not guarantee Google's chosen URL, indexing or ranking.
Canonicalization is not a substitute for a useful page and original evidence.

Google Search Central

Your choices / Not algorithm instructions

What to make

Write the search intent and related questions before drafting a descriptive title and answer-first outline.
Supply your own demonstration, experience or clearly sourced evidence; make authorship and limits visible.
Keep each page's purpose clear and use consistent canonical signals for actual duplicates rather than producing thin near-copies.

An outcome to define / Not a score

Search click-through rate

Count this outcome: Search clicks for the specified page and query cohort
Out of this eligible group: Search impressions for that same page, query cohort and window

Use consistent Search Console filters for search type, country and device when available. Aggregate CTR is confounded by query mix, rank and search features; a before/after title comparison is not a randomized causal test.

Quality & trust

What must not get worse

A visitor can accomplish the stated intent without returning just to find a missing basic answer.
Check factual accuracy, current evidence, clear limitations and useful downstream action.
Avoid misleading titles, thin duplicates and content written mainly to manipulate ranking.

Confounders

What else might explain it

Rank, query mix, device, geography and search features
Indexing, canonical selection and title rewrites
Seasonality, demand and competitor changes
Algorithm updates and delayed reporting

Source dates, access & references

Google Search Central
Creating helpful, reliable, people-first content (opens in a new tab)
Published: date not established · Editorial review: 2026-09-08
Guidance for assessing useful, reliable content made for people, including original information, clear sourcing and a satisfying answer. Not a checklist of directly measurable ranking weights.
Access & limits. Living documentation reviewed through the approved reference, not newly retrieved. The guidance does not reveal a complete retrieval or ranking algorithm, keyword volumes, or guaranteed positions. E-E-A-T is not a single ranking factor.
Google Search Central
What is URL canonicalization (opens in a new tab)
Published: date not established · Editorial review: 2026-09-08
How Google identifies a representative URL among duplicate or very similar pages. Canonicalization concerns indexing and consolidation, not emotional appeal.
Access & limits. Living documentation reviewed through the approved reference, not newly retrieved. A declared canonical is a signal, not a guarantee of Google's selected canonical or of indexing and ranking.

07 /

Strategy channel · not a ranking system

Marketing email

Permission-based email campaign

A strategy channel for a truthful promise, useful evidence and a clear next step. It has no single official ranking formula in this catalogue.

Channel boundary. Strategy channel, not an official ranking system. Inbox providers have distinct delivery and filtering systems; no provider-specific delivery, inbox placement or ranking formula is claimed here.

Read the strategy hypotheses

A specific subject-line promise aligned with the body is a candidate for testing useful response. Headline research can inspire a question, but cannot predict an email lift.

Context. Editorial transfer hypothesis for an opted-in, nonpolitical email audience.

What this does not establish

News-headline tests are not email experiments, and headline assignment does not isolate word attributes.
List composition and delivery can change who has a chance to respond.

Verified experiences and genuine deadlines can clarify an offer when their source, scope and terms are supplied. Whether their presentation improves useful response is a local hypothesis.

Context. An optional offer-framing experiment, not a disclosed inbox-ranking signal.

What this does not establish

The music-market and older scarcity studies do not establish email conversion effects.
Never generate testimonials, availability counts or deadlines; missing evidence remains an unfilled requirement.

Your choices / Not algorithm instructions

What to make

Write a specific promise, body evidence, all material terms and one clear CTA for the stated audience.
If testing a subject line, hold sender, preview text, body, offer and destination fixed.
Use consented recipients and an easy unsubscribe; do not optimize shame, fake scarcity or hidden costs.

An outcome to define / Not a score

Unique-recipient click rate

Count this outcome: Delivered recipients who clicked the intended CTA at least once, excluding identified automated clicks
Out of this eligible group: Recipients with a delivered message in the same assigned variant and window

Deduplicate by recipient and document bot filtering; security scanners can inflate clicks. Opens are not a reliable proxy for reading. Compare delivery rates too: conditioning only on delivered mail can break a randomized-group comparison if delivery differs.

Quality & trust

What must not get worse

Record unsubscribes, spam complaints and negative replies alongside useful follow-through.
The destination fulfills the subject-line promise and displays the complete offer.
Review delivery differences and whether clickers actually understood the terms.

Confounders

What else might explain it

Consent history, list source and audience mix
Deliverability, automated clicks and privacy protections
Send time, prior contact and concurrent campaigns
Offer availability or destination changes

Source dates, access & references

Aubin Le Quéré & Matias · Scientific Reports
When curiosity gaps backfire: effects of headline concreteness on information selection decisions (opens in a new tab)
Published: 2025 · Editorial review: 2026-09-08
Research on headline concreteness and information selection using online headline tests. Random assignment concerns complete headline alternatives, not an independently manipulated word attribute.
Access & limits. Editorial review uses the approved research references; the publisher page was not re-retrieved on the review date. Do not treat a headline-feature association as an isolated causal effect or extrapolate it to every platform.
Robertson et al. · Nature Human Behaviour
Negativity drives online news consumption (opens in a new tab)
Published: 2023 · Editorial review: 2026-09-08
Headline-test data from Upworthy and analysis of negative language in relation to click-through. Readers were randomly assigned whole headline alternatives in that publisher's testing environment.
Access & limits. Editorial review uses the approved research references; no fresh full-text retrieval. Random assignment of headlines does not independently isolate negative words from other headline differences. Click-through is not trust, satisfaction, learning or sales.
Salganik, Dodds & Watts · Science
Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market (opens in a new tab)
Published: 2006 · Editorial review: 2026-09-08
An artificial music market compared independent choice with conditions exposing participants to other listeners' choices. Social influence changed the distribution and predictability of success.
Access & limits. Editorial review uses the approved research references; no fresh full-text retrieval. A controlled music-market result does not establish that a review badge improves conversion, that popularity measures quality, or that any individual item will benefit.
Verhallen & Robben · Journal of Economic Psychology
Scarcity and preference: An experiment on unavailability and product evaluation (opens in a new tab)
Published: 1994 · Editorial review: 2026-09-08
A small, older experiment on availability explanations and product evaluation. The reason an item is unavailable matters to interpreting scarcity.
Access & limits. Editorial review uses the approved research references; no fresh full-text retrieval. The study is not a modern ecommerce field trial, does not establish a universal sales lift and does not validate fake timers or invented stock counts.

08 /

Strategy channel · not a ranking system

Marketing offer

Offer or landing page

Present value, evidence, price and constraints so someone can make an informed decision. This is a strategy channel, not an official ranking algorithm.

Channel boundary. No single official rank formula. This entry proposes testable presentation choices on a page you control; it does not claim that urgency, reviews or conversion rate are search or social ranking factors.

Read the strategy hypotheses

Showing one relevant, verified experience near the offer may help people assess fit, but the music-market social-influence experiment does not establish a landing-page conversion effect.

Context. A proposed evidence-presentation comparison on a nonpolitical product or service page.

What this does not establish

A testimonial must be real, attributable, permitted and relevant, not generated proof.
A higher conversion rate does not establish that buyers made better-informed decisions.

Salganik, Dodds & Watts · Science (2006)