The recorded ChatGPT responses are preserved. The context boxes and summaries explain how to read them.
Some code blocks remain unclassified. The labels describe the evidence shown in this session; they do not verify private system internals.
Ranking signals
What Ranking Information Did ChatGPT Report Seeing?
The session describes result order without exposed numeric ranking scores.
We ask whether a result card includes PageRank or other numeric ranking scores. ChatGPT says those scores are not exposed in the interface used for this session.
The reply concerns what the model can inspect. It does not establish which signals the search provider uses to select and order results.
What the response describes
ChatGPT lists titles, URLs, snippets, reference IDs, dates, result order, and source word limits. It reports no exposed PageRank, domain-authority value, or backlink count.
Result order shows where a page appeared in the returned list. It does not explain the provider’s scoring formula. The transcript cannot establish whether every score mentioned in the question exists in the underlying system.
Source word limits
The reply describes a word limit as a restriction on how much source-derived material an answer may use. A value of 200 is not a page-length measurement or a ranking score.
Dates
A publication date and a crawl date describe different events. Neither alone establishes that a price, specification, or other claim remains current. Check the relevant statement on the source page.
Result types
The response groups search results, opened documents, dedicated data lookups, and display formats. It lists 16 types under its initial classification. Treat that list as the session’s description, not a permanent inventory of every ChatGPT tool.
For the result card, does it contain any intrinsic values like PageRank? Or is it all text apart from the reference ID?
Also give me a list of every result type a result card might be in your process.
Does a result card contain PageRank or similar intrinsic scores?
No PageRank-like value is exposed to me. I do not receive:
- Google PageRank
- A generic authority score
- Domain authority
- Backlink counts
- Trust score
- Spam score
- Relevance score
- Semantic-similarity score
- Search-engine confidence
- Click-through rate
- Popularity score
- A numeric ranking score
The underlying search system almost certainly uses internal ranking signals to decide which results to return and in what order, but those signals are not included in the result object I can inspect. I see the ordering of the results, but not the numerical reason for that ordering.
So I may know:
This result appeared first.
I do not know:
This result received a relevance score of 0.913, has a PageRank of 7.4, or was ranked first because of 42 specific signals.
The ordering itself is the only visible ranking signal, and even that should be treated cautiously. First position means the retrieval system ranked it highly for that query; it does not necessarily mean it is the most authoritative or best evidentiary source.
It is not entirely unstructured text
A typical web-search result contains structured metadata in addition to text. The exact fields vary, but I may receive:
| Field | What it represents |
|---|---|
| Reference ID | Internal identifier used to open or cite the result |
| Title | Page title associated with the result |
| URL | Destination or canonical-looking URL |
| Snippet/extract | Query-related text preview |
| Publisher/domain | Usually inferable from the URL and sometimes separately represented |
| Publication date | When detected and available |
| Crawl date/freshness | When the search index or retrieval system last processed it |
| Result order | Its position within the returned results |
| Content-type indication | Sometimes identifies a PDF, news item, image, or other special result |
| Image indicator or thumbnail | Sometimes present when the page has an associated image |
| Word-limit metadata | An internal limit on how much source-derived material may be used in the answer |
Example structure
A result may effectively appear to me as something resembling:
Title: How People Are Using ChatGPT
URL: https://openai.com/index/how-people-are-using-chatgpt/
Reference ID: turn123search4
Published: September 15, 2025
Crawled: 3 days ago
Snippet: OpenAI analyzed 1.5 million consumer conversations...
Word limit: 200That is a conceptual representation, not necessarily the literal internal serialization.
What the word limit means
The word-limit value is not a quality score and is not a property of the webpage itself.
It is an instruction governing how many words in my response may be attributed to or derived from that source. It helps control source overuse and copyright-related summarization limits.
It does not mean:
- The article has only 200 words.
- Only 200 words were indexed.
- The page is less important.
- The search system is 200 units confident in it.
Publication date versus crawl date
These are distinct:
- Publication date: when the publisher says the article or page was published.
- Crawl date: when the search or retrieval system most recently accessed or indexed it.
Neither is always available or completely reliable.
A publication date can be:
- Missing
- Incorrectly inferred
- Changed when an article is updated
- Confused with an HTML modification date
- Pulled from structured metadata that the publisher configured incorrectly
A crawl date only indicates retrieval freshness. It does not prove the page’s underlying information is current.
What the snippet is
The snippet is primarily textual, but it is not necessarily a publisher-written summary. Depending on the result, it may be:
- A search-index extract
- Text surrounding relevant query terms
- The page’s metadata description
- A generated condensation
- A fragment from the opening paragraphs
- A mixture of indexed text and metadata
I am generally not shown a reliable provenance label explaining exactly how a particular snippet was produced.
That is why the snippet is used for triage, not as definitive evidence.
Result types in the research process
There are several layers here. Some are true search-result cards; others are structured tool results or opened-source objects. It is useful to keep them separate.
1. General web-search result
This is the standard result card.
Typical targets include:
- HTML webpages
- Company pages
- Blog posts
- Government pages
- Academic landing pages
- Research-center pages
- Documentation
- Forums
- Social or community pages
- Dataset descriptions
- Press releases
Typical fields:
- Title
- URL
- Snippet
- Reference ID
- Publication date, occasionally
- Crawl date
- Word limit
- Sometimes an image indicator
The internal reference usually resembles:
turn123search4The word search identifies it as a normal search result.
2. News result
A news result is associated with current or time-sensitive journalism.
It may contain:
- Headline
- Publisher
- URL
- Publication time or date
- Snippet
- Reference ID
- Possibly an image
Its reference identifier may resemble:
turn123news4News results can be displayed in a group of source links, but they still need ordinary source evaluation. Being classified as news does not establish quality.
Possible news publishers include:
- Newspapers
- Wire services
- Television news sites
- Trade publications
- Technology publications
- Local news outlets
- Corporate newsrooms
The distinction from a general web result is partly related to recency and publisher classification.
3. PDF result
A PDF can initially appear as a regular search result that points to a PDF URL.
Examples include:
- Research papers
- Survey reports
- White papers
- Government reports
- Technical specifications
- Conference papers
- Slide decks
- Annual reports
Once opened, it becomes a PDF-view object. At that point, I may receive:
- Parsed text
- Page count
- Section structure
- Links
- A PDF-specific reference
- The ability to render individual pages as screenshots
A PDF reference after opening may resemble a view object rather than a standard search result.
For PDFs, page screenshots are required when I need to inspect:
- Charts
- Tables
- Diagrams
- Page layout
- Images
- Figures not represented in parsed text
A PDF card does not expose citation count, journal impact factor, PageRank, or document-quality scores unless those happen to be written in the page text.
4. Image-search result
An image query returns image-specific results rather than ordinary webpage cards.
An image result may provide:
- Image preview
- Image source
- Host page
- Image URL or reference
- Caption or title
- Dimensions or other image metadata, when available
- Internal image reference ID
Its reference may resemble:
turn123image4Image results are primarily used when visual material adds value, such as for:
- People
- Animals
- Locations
- Historical events
- Architecture
- Products
- Visual comparisons
An image result can be displayed in an image carousel.
These results do not expose image-search ranking scores or conventional PageRank.
5. Product-search result
A product query returns a retail-product result with structured commerce information.
Possible fields include:
- Product name
- Image
- Price
- Retailer
- Availability
- Product variant
- Rating or review information
- Merchant link
- Product reference ID
Its identifier may resemble:
turn123product4Product results are different from ordinary webpages because the search system attempts to identify the product as an entity and consolidate commerce-related metadata.
They are used for:
- Product recommendations
- Price comparisons
- Shopping research
- Looking up specific retail products
- Comparing models or variants
They may contain price or rating values, but those are product attributes, not search-ranking scores.
6. Opened webpage view
When I open a regular search result, the result becomes an expanded page view.
The opened page may provide:
- Extracted page text
- Headings
- Paragraphs
- Links with clickable IDs
- Relevant passage positioning
- Publication metadata
- URL
- Content type
- Internal view reference
A view reference may resemble:
turn123view4The opened view is not another independent search result. It is the expanded representation of a result or URL.
It allows operations such as:
- Find a word or phrase
- Open a particular location in the text
- Follow links found on the page
- Read more complete context
- Determine whether the result actually supports a claim
7. Opened PDF view
This is a specialized form of opened-source object.
In addition to extracted text, it permits:
- Page-by-page navigation
- PDF screenshots
- Examination of charts and tables
- Visual interpretation of figures
- Verification of page-specific content
A PDF view is especially important because parsed PDF text can scramble:
- Column order
- Table cells
- Chart labels
- Footnotes
- Multi-panel figures
8. Finance result
A finance lookup produces structured market data rather than a normal webpage result.
Possible fields include:
- Ticker
- Instrument name
- Current price
- Price change
- Percentage change
- Market
- Trading status
- Historical price series
- Currency
Reference identifiers may resemble:
turn123finance4Supported instrument categories can include:
- Public equities
- Funds
- Cryptocurrencies
- Market indexes
These results can be displayed as an interactive price chart.
This is not an ordinary URL discovered through web search. It is a dedicated data result.
9. Weather result
A weather lookup produces structured forecast data.
Possible fields include:
- Location
- Current conditions
- Temperature
- Daily highs and lows
- Precipitation
- Wind
- Forecast period
- Hourly or daily forecast
Its reference may resemble:
turn123forecast4This can be displayed as a forecast widget.
Again, it is not a conventional search-result URL.
10. Sports schedule result
A sports schedule lookup may contain:
- Team
- Opponent
- Date and time
- Venue
- Game status
- Score
- Broadcast information
- Upcoming or previous games
Its reference may resemble:
turn123sports4This can be rendered as a schedule or live-score display.
11. Sports standings result
This uses the same broad sports-result family but contains standings rather than games.
Possible data includes:
- Team
- Wins
- Losses
- Ties
- Winning percentage
- Games behind
- Division or conference position
- Other league-specific statistics
It may use a sports reference ID like:
turn123sports4Whether it becomes a schedule or standings display depends on which sports operation produced it.
12. Current-time result
A time lookup produces:
- Current local time
- UTC offset
- Date
- Sometimes timezone context
Its reference may resemble:
turn123time4This is structured utility output, not a webpage result.
13. Calculator result
A calculator operation returns:
- The expression
- The calculated value
- Optional unit, prefix, or suffix
It does not generally need a web citation because it is a deterministic calculation rather than external evidence.
This is not a search-result card in the strict sense.
14. Map or local result
ChatGPT Search can display map-oriented results in some user-interface contexts, particularly for local queries. Official OpenAI documentation notes that mobile search responses may show a map when appropriate. (OpenAI Help Center)
However, in the research interface used for the previous project, I did not have a separately exposed general local-business or map-search result type comparable to the dedicated weather, product, or sports operations.
Therefore:
- Map results exist in some ChatGPT Search experiences.
- I should not claim that I had map-result cards available in that particular research run.
15. Citation/source annotation
A citation is not itself a search result. It is a reference in the generated answer pointing to an opened or otherwise used source.
It normally includes:
- Internal source reference
- Link to the source
- Enough metadata for the user interface to display the source title or preview
Official OpenAI documentation describes web citations as references to web resources used to generate the response. (OpenAI Platform)
A result can be:
- Discovered but never cited
- Opened but never cited
- Used for internal context but not cited
- Directly cited in the answer
16. Search-results navigation group
For recent news, multiple news results can be assembled into a visible set of links. This is a presentation object rather than a new underlying source type.
The group may contain several news references.
Similarly, the interface may present:
- An image carousel
- A product carousel
- A finance chart
- A weather forecast
- A sports schedule
- A standings table
These are rendering formats, not additional discovery-result categories.
Condensed classification
Actual discovery cards
- General web result
- News result
- PDF/document result
- Image result
- Product result
Expanded source representations
- Opened webpage view
- Opened PDF view
Dedicated structured-data results
- Finance result
- Weather forecast
- Sports schedule
- Sports standings
- Current time
- Calculator
Interface-dependent result presentation
- Map/local display
- News-source group
- Image carousel
- Product carousel
- Finance chart
- Weather widget
- Sports schedule widget
- Sports standings widget
Only the first five are reasonably described as search-result cards. The rest are expanded resources, direct data responses, or ways of displaying results.
Bottom line
A standard result card is primarily:
Structured bibliographic metadata plus a textual relevance preview.
It does not expose quantitative authority or ranking signals. The most important nontextual or structured values I ordinarily see are:
- Reference ID
- URL
- Result position
- Publication date
- Crawl date
- Content type
- Occasionally image-related metadata
- Internal source word limit
The absence of a PageRank-like score is why source evaluation must be based on provenance, methodology, relevance, freshness, and direct inspection—not on an exposed authority metric.