SERP Research Methodology.
A search engine results page is a time-bound, context-dependent observation surface. SERP research measures what appeared for a declared query, engine, location, language, device, depth and timestamp—then separates that observation from any explanation of why it appeared.
SERP research observes result states—not ranking causes.
The same query can produce different surfaces across location, language, device, time and interface. A defensible study records those coordinates before aggregating positions into domain-level patterns.
SERP research methodology is the controlled observation, classification and comparison of search-result surfaces across a versioned query corpus under declared contextual and temporal conditions.
Organic results, ads, local modules, related questions, images, video and answer features occupy different search real estate.
Layout, feature occupancy, pixel depth and result type affect what a user can encounter.
Content, links, history, intent fit and system processes are hypotheses requiring separate evidence.
One SERP observation requires eight coordinates.
Remove any coordinate and later comparisons may combine states that were never equivalent.
Exact query text
Preserve normalized query ID and original submitted text. Similar phrases are separate units until a declared grouping rule combines them.
Search engine and type
Declare the engine, vertical and requested result mode. Web, image, news and maps are not interchangeable populations.
Geographic context
Use an explicit location identifier or validated coordinate rule. Country-level and city-level observations answer different questions.
Language context
Preserve requested language and corpus language. Language changes interpretation, result supply and query meaning.
Device state
Desktop and mobile may differ in composition, layout and visible depth. Never pool them silently.
Requested result depth
Fix the maximum observed range and distinguish valid absence within depth from data that was never requested.
Collection timestamp
Preserve capture, retrieval and snapshot times. Asynchronous collection must not be presented as one simultaneous SERP.
Acquisition and response state
Record the collection method, run identifier, status, raw observation and failures before applying normalization or derived metrics.
Rank order and surface occupancy are different measurements.
A domain can hold a strong organic position while appearing below several feature modules. The protocol should preserve both ordered organic position and broader SERP composition.
Preserve the ordered list and the surface map.
Ordered organic position supports ranking distributions and domain share. Surface composition supports feature prevalence and result-type research. Pixel-based visibility requires rendered-page measurement and should not be fabricated from rank alone.
Choose the design before collecting SERPs.
A snapshot, matched comparison, volatility study and feature study use the same raw environment differently. The console changes the unit, repetition, output and claim boundary.
Describe one synchronized result state
Make every aggregation formula-visible.
The following are transparent analytical constructs calculated from validated observations. They are not presented as search-engine authority scores or facts supplied by the acquisition system.
Query presence rate
queries with ≥1 domain result ÷ valid queriesMeasures how often a normalized domain appears at least once within declared organic depth.
Organic result share
domain organic results ÷ all valid organic resultsUseful when multiple URLs from one domain can appear. Must declare deduplication and depth rules.
Position-weighted share
Σ declared weight(position) ÷ Σ all weightsThe weighting function must be published. It expresses an analytical preference, not measured click behavior unless calibrated separately.
Result-set overlap
|A ∩ B| ÷ |A ∪ B|Jaccard overlap can compare normalized domain or URL sets across queries, markets, devices or time.
Persistence rate
units present in every wave ÷ units present in any waveOne transparent stability view. Other definitions are valid when explicitly versioned and justified.
Completion rate
valid query snapshots ÷ planned query snapshotsAlways report before other aggregates. Low completion can distort every downstream comparison.
Precise examples with explicit inference limits.
All numbers below are synthetic demonstrations of method logic—not live SERP data, source-system claims or performance statements about any domain.
SERP datasets fail through silent context loss.
These errors can leave a technically complete table while destroying comparability or overstating the conclusion.
Query drift
The corpus changes between waves without a version boundary, making trend claims depend on different questions.
Context mixing
Locations, languages, devices or engine types are pooled as if they were one observation environment.
Feature collapse
Organic results and heterogeneous modules are treated as identical ordered positions.
Domain duplication
Subdomains, protocols, host variants or tracking URLs split one identity or merge distinct properties.
Temporal smearing
A long asynchronous collection window is labeled as one instantaneous ranking state.
Silent task failure
Only successful query responses remain, changing the denominator and possibly the topic distribution.
False causality
A ranking change is attributed to one page edit, link or update without a design that tests alternatives.
Metric opacity
A proprietary-looking visibility score hides depth, weights, denominator and missing-data treatment.
A ranking table needs an observation record.
The minimum manifest preserves enough context to reconstruct what was requested, what returned, what was classified and what was derived.
Every method node. One controlled research route.
MTH/05 specializes the general acquisition controls for search results. MTH/06 applies the same evidence discipline to entities, attributes and relations.