# Separate Original Reports From Syndicated Copies in Web Research

Several search results do not, by themselves, give independent evidence for a claim. Different URLs can repeat one original source. Read the relevant passages, follow their citations, and record which pages depend on the same report. Find the earliest attributable source you can locate. If the source chain is missing, mark independence as unknown. An earlier visible date does not prove authorship.

Start with a manual source check. Open the report, announcement, or study named by each page. Then compare what each source actually contributes. Search tools can help you find candidate pages. You must still check the links between their claims and their evidence.

The result is a source-dependency table: a record that links each precise claim to its sources and shows which sources repeat others. It helps you separate one announcement with several copies from several sources with their own evidence.

## Define the claim before you collect more pages

A public buyer asked for “quick competitor research” and wanted to avoid invented answers and basic comparisons. That wording points to a useful starting rule: choose a specific comparison question before collecting more results. The post gives context for this guide. It does not show how common the problem is. [Public research question — published August 12, 2026; observed October 25, 2026; individual wording only](https://www.reddit.com/r/AskMarketing/comments/1vmjrrb/how_are_you_doing_competitor_research_with_ai/).

Write one claim that you can check. Include the product, capability, conditions, and relevant time or version when known. Avoid a broad statement such as “this product is better.” That statement hides several questions about features, test conditions, and evidence.

Separate these two questions:

1. Did the company announce the capability?
2. Does independent evidence show that the capability works under the stated conditions?

An official announcement can directly support the first question. Several articles that repeat it do not automatically answer the second.

Define what evidence you need before you search. For an announcement claim, the official announcement may be enough. For a performance claim, seek a separate test with relevant methods and conditions. This keeps you from collecting more copies of evidence that cannot answer your question.

## Use search to find candidates

An Actor is a hosted data collection program. A Task is a saved input preset for an Actor. The [Google Search Scraper Actor](https://apify.com/chronometrica/google-search-scraper) fits the candidate-finding part of this method. Its documented search fields do not establish whether two reports have independent evidence.

The checked public default build, version `1.0.15`, declares the input fields below. A static check passed: the proposed input matched the declared input rules without running the Actor. This does not establish live results, complete coverage, or successful collection. [Public default build — checked October 25, 2026; declared fields and static validation only](https://api.apify.com/v2/acts/mlmVX4amk2biciJaB/builds/default).

Use this small proposed input for an initial search:

```json
{"country":"US","engines":["google"],"language":"en","maxResultsPerExecutedQuery":3,"queries":["reporting software official announcement"]}
```

Use search to find candidates

Replace the complete sample query list with your intended search phrase. Review the complete engine list, country, language, and result limit before any run. Do not assume a saved Task already contains these values.

The sample requests Google, US country settings, and English language settings. `maxResultsPerExecutedQuery` sets a maximum of three results per executed query. Three is a limit, not a promise of three returned pages. This is proposed input. No Actor run occurred for this guide.

The [Competitor SERP Research Task — checked October 25, 2026; public page presence only](https://apify.com/chronometrica/google-search-scraper/examples/competitor-serp-research) is a starting point. SERP means search engine results page. The public page contains Task ID `Pjq0JLkzo0wHTcabU`. Its current saved settings and runtime behavior remain unverified.

Keep public evidence separate from account records. Public Actor documentation and a public Task page do not prove your saved account settings or run outcomes. Record those separately if you later use the Actor. Do not present the proposed input as a completed collection.

## Save search context and source relationships separately

Keep the search context so you can explain how you found a page. Then create a separate table for your source decisions. A search result describes a discovery. Your source table records what you checked after discovery.

The checked build declares these output fields. Their presence in the declaration does not prove that a particular run returned them. [Public default build — checked October 25, 2026; output declaration only](https://api.apify.com/v2/acts/mlmVX4amk2biciJaB/builds/default).

| Search field | What to retain it for |
| --- | --- |
| `engine` | The search engine used |
| `executedQuery` | The executed search wording |
| `rank` | The result position reported by the Actor |
| `title` | The search result title |
| `canonicalUrl` | The returned candidate address |
| `resultDomain` | The reported result domain |
| `snippet` | The search text, which may be null |
| `isSponsored` | The reported sponsored-result flag |
| `observedAt` | The collection observation time |

A snippet is text from the search result. It is not a checked article passage. A result address does not identify the original author. Search rank does not measure evidence quality. Keep `observedAt` separate from any publication date shown on the page.

A null `snippet` means the value is missing. It does not mean the article contains zero relevant words. If a publication date is absent, record it as unknown. Do not replace it with the observation time.

Repeated export rows are a separate problem. Use the [search export merging guide](/en/articles/web-result-dedup/) if you need to preserve search observations while handling repeated addresses. Here, keep different pages when they help show how a claim spread.

Build your own review table with these fields. These are proposed analysis fields, not additional Actor output fields.

| Review field | What to save |
| --- | --- |
| Claim ID and exact claim | The statement under review |
| Page ID and URL | A stable reference to each checked page |
| Publisher and named author | The identities shown, or unknown |
| Page date or version | The displayed value and its stated meaning |
| Review time | When you checked the page |
| Passage and locator | A short relevant passage and its heading or location |
| Cited source URL | The reference supporting this particular claim |
| Parent source ID | The source this passage depends on, if established |
| Relationship evidence | Citation, attribution, or matching passage |
| Source group | The shared source behind related passages |
| Independent contribution | Any separate observation, test, or analysis found |
| Status and next action | Your decision, remaining gap, and next check |

A parent source is the source another page uses for the claim. A source group contains the pages linked to that same evidence. Leave the parent unknown when you cannot establish the relationship.

## Follow the claim through its source chain

Use these eight steps for each claim. Keep the scope narrow enough that you can read every relevant passage.

1. Write the exact claim. State what must be established. Separate an announcement from a claim that the announced capability works.
2. Collect a bounded set of candidates. Save the query, settings, and returned addresses. A bounded set has a stated limit. It is not a complete survey of the web.
3. Read the relevant passages. Check the actual page and the source attached to the claim. A general reference list may not show which source supports which sentence.
4. Follow the references. Open each cited report or announcement. Continue through intermediate summaries when needed. Record every established connection.
5. Locate an attributable primary source. A primary source supplies the announcement, observation, or test being discussed. Record the earliest attributable source you located, with the limits of your search.
6. Group dependent passages. Put explicit summaries and copies under the source they use. Preserve their separate URLs and relationship evidence.
7. Assess separate contributions. Keep genuinely separate observations in their own groups. Check what each observation establishes and under which conditions.
8. Record gaps and write the conclusion. Leave missing source chains unresolved. Report page counts, source roles, and independent support separately.

For the passage check in step three, use the [guide to checked page evidence](/en/articles/web-agent-evidence/). That method helps you connect a claim to actual page text. This guide adds the next question: does that text rely on evidence already represented elsewhere?

Follow the source for the specific claim. A page may link to an official announcement for a feature statement and contain separate analysis elsewhere. The announcement link does not make every sentence on the page dependent. The separate analysis also does not make the repeated feature statement independent.

Stop a source chain when you reach the relevant primary evidence or when the next connection cannot be checked. Record the reason for stopping. Do not silently jump from a summary to an assumed original.

## Decide whether the evidence is independent

Independence concerns the evidence behind the claim. Different authors, domains, or titles are not enough. Two publishers can both use one announcement. One page can also contain both a repeated claim and a separate test.

Use an explicit citation as evidence of a relationship. Check that it refers to the passage under review. Save the cited address and the wording that links the claim to it.

Matching passages need more care. A distinctive repeated sentence can suggest copying or a shared source. It may not establish who copied whom. If the page gives no attribution and you cannot resolve the relationship, record the match and mark the dependency as unclear.

Treat dates as clues. An earlier displayed date can help direct your next check. It does not prove that the page created the claim. Record what the date represents when the page makes that clear. Keep an unknown date unknown.

Use these review states consistently:

| State | Meaning |
| --- | --- |
| Located primary source | A checked source supplies the relevant original announcement, observation, or test |
| Attributed repeat | The passage explicitly relies on a located source |
| Dependency unclear | A possible relationship exists, but the chain is unresolved |
| Separate observation located | The passage supplies a checked observation distinct from the shared source |
| Unavailable | You could not check the required page or passage |

A separate observation still needs a quality check. Read its methods, scope, and conditions. Independence does not make a weak test strong. A test of one version or condition may not support a broader claim.

Do not infer an absent capability from absent documentation. A checked page may fail to support a feature claim without proving that the feature does not exist. Record the narrower finding: the checked evidence does not establish it.

## Original fiction: five pages repeat one announcement

This is an original fiction teaching example. It is not an Actor run, customer result, or observed company finding.

A researcher checks the fictional claim: “The product supports scheduled report delivery.” Five pages discuss it. One is the official announcement. Three explicitly cite that announcement. The fifth repeats one sentence without naming a source.

| Page | Checked evidence | Source relationship | Decision |
| --- | --- | --- | --- |
| A | Official announcement states the capability | Located primary announcement | Start source group A |
| B | Relevant passage cites A | Attributed repeat of A | Place under group A |
| C | Relevant passage cites A | Attributed repeat of A | Place under group A |
| D | Relevant passage cites A | Attributed repeat of A | Place under group A |
| E | Matching sentence with no source | Dependency unclear | Keep unresolved |

The page accounting is:

1 announcement page + 3 attributed repeat pages + 1 unclear page = 5 pages reviewed.

The established source group contains:

1 announcement page + 3 attributed repeat pages = 4 pages linked to one announcement.

Page E remains outside the established chain. Its matching sentence is a reason to investigate. It is not enough to assign a parent source with certainty.

The evidence summary is one located primary announcement, three attributed repeats, and one unclear source. There are 0 separate independent tests located among these five pages. That zero describes the reviewed set. It does not mean no independent test exists anywhere. Page E has an unknown source relationship, not a zero-valued relationship.

The supported conclusion is: “The official announcement states that scheduled report delivery is supported. Three checked pages repeat that announcement. One further page has an unresolved source relationship.”

Do not turn this into “five independent confirmations.” Do not call the capability independently verified. Seek a separate test that checks the capability under relevant conditions. If you locate one, record its methods and results before changing the conclusion.

## Check gaps before you report a result

Review the complete table before you write the comparison. Check both the source relationships and the wording of the final claim.

1. Confirm that every counted page has an ID and a recorded review state.
2. Confirm that every assigned parent has relationship evidence.
3. Check that one shared report was not counted again through its summaries.
4. Keep unavailable pages separate from pages you read but found insufficient.
5. Preserve missing values and distinguish them from observed zero counts.
6. Compare the final claim with the source conditions, dates, and versions.
7. State the search limit and unresolved source chains.

If a cited page is unavailable, retain the citation and mark the passage unchecked. Do not treat the citation as verified support. This guide does not promise recovery of historical content. Use an older retained source only if you already have it, and identify that version.

If the trail ends at a summary, name it as the earliest located source rather than a proved original. If several pages appear connected but no chain can be established, report unknown independence. Do not use uncertainty to increase the confirmation count.

The final report should answer three questions: what was located, how the sources relate, and what remains unproved. State the number of reviewed pages separately from the number and quality of separate evidence sources.

## Choose the next source check

Start with one claim and complete its source-dependency table. If you need to establish what a company announced, check its official source first. If you need to establish whether the capability works, seek a separate relevant test. If the source chain is unresolved, make that the next research task.

Use the [Google Search Scraper Actor](https://apify.com/chronometrica/google-search-scraper) when you need a bounded set of candidate pages. Review the proposed input before a run. Then inspect the passages and source relationships yourself. The next decision should follow the evidence gap, not the number of repeated results.

## Sources

- [Google Search Scraper public default build — checked October 25, 2026; build 1.0.15 declarations and static input validation only, no live run](https://api.apify.com/v2/acts/mlmVX4amk2biciJaB/builds/default)
- [Competitor SERP Research Task — checked October 25, 2026; public page and exact Task ID confirmed, saved settings and runtime unverified](https://apify.com/chronometrica/google-search-scraper/examples/competitor-serp-research)
- [Public competitor research question — published August 12, 2026; observed October 25, 2026; individual wording, not measured demand or product evidence](https://www.reddit.com/r/AskMarketing/comments/1vmjrrb/how_are_you_doing_competitor_research_with_ai/)