Being on the page AI reads: 5.28M pairs and the 34.0% that got named
34.0% of 5.28M measured page-brand-answer pairs were named. Compare engines, page shapes, prominence, and frequency.
Originally published October 7, 2026
The answer named the brand in 34.0% of measured page-brand-answer pairs. Citation of a page did not guarantee naming every brand on it.
September 7–October 6, 2026 (UTC) · 5,280,084 pairs; 275,150 pages; 101,938 brands
- Named in the answer
- 34.0%1,793,590 of 5,280,084 pairs
- Measured pages
- 275,150Distinct cited sources
- Detected brands
- 101,938Not independent trials
Being present on a cited page often did not mean being named in the answer. In 5,280,084 measured page-brand-answer pairs, the answer named the brand 34.0% of the time. The result separates a placement's presence in a source from the outcome a marketer usually wants: the brand appearing in the answer itself.
The measurements come from Parse, the AI answer index Dimitry Apollonsky also runs; the analysis and interpretation are Signals'. This is an observational study, not a test of a Signals service.
The window is September 7 through October 6, 2026, covering 275,150 distinct measured pages and 101,938 brands. Signals runs an aged Reddit account marketplace plus an editorial network for AI brand mentions across Reddit, Quora, Product Hunt, and Threads.
The answer named about a third of measured pairs
The headline counts 1,793,590 named pairs among 5,280,084 eligible pairs. Each pair connects a detected brand on a cited page with the answer that cited that page. The naming rate is therefore conditional on citation and measured page content; it is not the probability that publishing a mention makes the brand appear.
A page containing several brands contributes several pairs. If the same page is cited in several answers, those answers contribute separate observations. This weighting reflects repeated use of the source, but it also means the observations are not independent trials of a placement strategy or a set of unique brand purchases.
The practical implication is to verify the actual answer after a source appears. A cited URL can be useful evidence while most of its listed brands remain unnamed. The headline should travel with that denominator, especially when it is quoted in a placement brief or used to assess a vendor's results.
The engine split changes the baseline
ChatGPT Search named the brand in 40.13% of its measured pairs, compared with 27.70% for Google AI Mode. The chart uses each engine's eligible page-brand-answer pairs as its own denominator. These are observed source sets, not a matched experiment giving both engines the same pages and requiring the same answer format.
The pair-level naming rate differed by engine
Pairs whose brand appeared in the answer · % · September 7–October 6, 2026 (UTC)
View data table
| ChatGPT Search | 40.13 |
|---|---|
| Google AI Mode | 27.7 |
ChatGPT Search: 1,068,110/2,661,410; Google AI Mode: 725,480/2,618,674. Source: Parse index, analyzed by Signals.
Observed engine source sets, not a matched-page experiment.
The difference may reflect the prompts answered, brands present, source selection, or the number of alternatives on the cited pages. It should not be interpreted as an isolated engine preference for a particular mention strategy. The same brand can also appear in many observations, contributing more weight than a rarely cited brand.
When planning a campaign, keep the engine in the reporting table. A blended result can conceal a change limited to one source set. Preserve the questions and full answers so a later comparison can distinguish altered coverage from a change in the brand's treatment on a stable panel.
Listicle-shaped pages had a lower pair-level naming rate
Brands on listicle-shaped cited pages were named in 27.94% of measured pairs, versus 39.03% on other page shapes. That is an observational comparison, and its direction matters: this cut does not support a blanket claim that appearing on a listicle makes a brand more likely to be named than appearing elsewhere.
Listicle-shaped pairs had a lower naming rate
Pairs whose brand appeared in the answer · % · September 7–October 6, 2026 (UTC)
View data table
| Other page shapes | 39.03 |
|---|---|
| Listicle-shaped | 27.94 |
Other: 1,119,792/2,868,914; Listicle-shaped: 673,798/2,411,170. Source: Parse index, analyzed by Signals.
A list may contain many brands that the answer does not select. Shape is heuristic.
A plausible contributor is the denominator. A roundup can list many alternatives while the answer selects only a few. However, the analysis does not establish that explanation as the cause of the difference. Page shape also varies with topic, publisher, brand, prominence, and the nature of the buyer question.
Shape comes from title and URL patterns rather than a manual review of every article. The listicle study describes that classification and a separate entry cohort. Use the figure to challenge an oversimplified placement pitch, not to conclude that every non-listicle page is a better editorial choice.
Primary mentions were named more often than passing mentions
The primary-mention group had a 47.68% naming rate, compared with 17.96% for passing mentions. These prominence labels describe detected page content. They are not experimental assignments, so the gap does not estimate what would happen if a publisher moved the same brand from a passing reference into a dedicated section.
Primary mentions had a higher observed naming rate
Pairs whose brand appeared in the answer · % · September 7–October 6, 2026 (UTC)
View data table
| Primary | 47.68 |
|---|---|
| Secondary | 29.22 |
| Passing | 17.96 |
primary: 1,007,539/2,113,285; secondary: 563,817/1,929,684; passing: 222,234/1,237,115. Source: Parse index, analyzed by Signals.
Prominence is a detected label; no randomized relocation of the mention occurred.
A page primarily about a familiar brand may also be more relevant to the question or offer clearer identifying information. A passing mention can appear in a long article about another product. Those differences are part of the observed groups and are not removed by placing their percentages side by side.
The useful editorial response is to request a meaningful explanation of fit, not a repeated name. Specify the use case, evidence, and limitation the passage should communicate. A prominent paragraph that says nothing concrete does not become a tested recommendation strategy merely because the primary group has a higher observed rate.
More detected mentions did not form a simple dose curve
The frequency groups describe how often a brand name was detected on the page, not how many mentions an operator should purchase or insert. The figure compares once, two to four, and five or more detected mentions. Their naming rates vary, but the groups differ in many ways beyond the number of repetitions.
Naming varied with detected mention frequency
Pairs whose brand appeared in the answer · % · September 7–October 6, 2026 (UTC)
View data table
| 2 to 4 | 24.88 |
|---|---|
| 5 or more | 44.57 |
| Once | 19.19 |
2 to 4: 382,901/1,539,059; 5 or more: 1,216,566/2,729,568; Once: 194,123/1,011,457. Source: Parse index, analyzed by Signals.
Frequency is correlated with page purpose and prominence. This is not a repetition prescription.
Repeated names may indicate a product-focused article, a long discussion, navigation text, or other page structures. A count does not establish the quality of the explanation. The analysis does not hold page length, topic, brand familiarity, or prominence constant, so it cannot isolate a repetition effect from those related characteristics.
For a brief, ask whether each additional mention carries useful information. A product description, a compatibility note, and a limitation can justify separate references. Repeating the same claim to reach a numeric quota cannot be justified by this study. The measurement is an observed grouping, not an instruction to optimize a keyword density.
Linked and unlinked mentions were not separated
This analysis does not identify whether the brand reference carried a hyperlink to the brand's website. The detected-name join supports a naming comparison, but it cannot establish an unlinked-mention advantage or quantify the effect of a backlink. Those questions require a text-and-link extraction that was not part of this measurement.
The distinction is important because the guide targets a question about unlinked mentions. We can explain what an unlinked mention is and how to write a useful passage without claiming this dataset settled the causal comparison. The brand-mentions guide keeps that boundary explicit when describing the vendor brief.
A future comparison would need to define which destination counts as a brand link, handle redirects and multiple brand domains, and inspect the link near the actual mention. It would also need comparable pages and brands. Counting all pages with any outgoing link would not answer the narrower question about the brand reference.
Latest page content is a retrospective measurement
We selected the latest available snapshot before October 7, 2026 for each eligible source page. That content may be newer than an answer within the observation window. A detected brand could have been added after the citation, while a removed passage could be absent from the selected snapshot despite existing when the engine answered.
The join therefore measures co-occurrence between an observed answer and a selected page snapshot, not a confirmed chain of reading and recommendation. It cannot tell us which cited source supplied a particular brand name when several sources were available. It also cannot observe what the engine retained or discarded internally.
Missing snapshots and missing detections select which pages enter the analysis. The 275,150 measured pages are not every page cited in the window. A refresh should report that selection again instead of assuming the measured subset represents all source material equally. The limitation belongs alongside the headline, not only in an unpublished analysis note.
The second count checked the aggregation
We calculated the headline directly over eligible page-brand-answer pairs and recomputed it by first grouping the same pairs by answer and observation time. Both returned 5,280,084 pairs and 1,793,590 named pairs. This verifies the aggregation by a different SQL formulation, while leaving shared source-selection and detection assumptions intact.
The receipts preserve the SQL, raw rows, and a number trace for each figure. The engine, shape, prominence, and frequency groups each reconcile to the same pair population. Percentages are calculated from their own group denominators and rounded for presentation; a missing measurement is not silently converted into zero.
That agreement does not replace a manual validation of every detected brand. Prominence remains a classifier label, and the study does not independently adjudicate every name or ambiguous product reference. Treat the analysis as a reproducible observed cut, with those limits, rather than a validated estimate of a universal probability for all brands and pages.
The operator should buy a concrete article, not a naming promise
The finding supports a more demanding placement brief. Specify what the article needs to explain and then inspect the answer separately after publication. A source URL appearing in an answer is useful evidence, but it does not establish that the brand was named, favorably described, or recommended for the relevant buying need.
Start with the brand-mentions guide for the host, passage, disclosure, and delivery requirements. Use Signals research to compare the other source types. Each article states its own denominator because citation occurrences, distinct pages, and page-brand-answer pairs answer different questions and should not be blended into one league table.
We do not have a complete blog-placement ledger or a controlled service test in this batch. The analysis can improve the questions an operator asks before buying. It cannot supply a guaranteed outcome, a measured return per placement, or a deadline by which an engine should change its answer.
Questions about the naming rate
The headline describes a conditional observation, so its safest use is to keep the condition visible. It concerns measured brands on pages already cited in the tracked answers. It does not start with all published mentions, and it does not establish which page caused an answer to name a brand when several sources were present.
Is this the chance a new brand mention gets cited?
No. The sample begins with cited pages and then checks whether detected brands appeared in the corresponding answers.
Does named mean recommended?
No. The matching rule checks brand naming, not a favorable recommendation or purchase endorsement. The wording of the answer still needs review.
Did Signals placements produce this result?
The result comes from Parse's observed source and answer data. It is not a measured effect of Signals placements.
How we measured
Read-only production queries frozen at October 7, 2026. Answers are ChatGPT Search and Google AI Mode in the stated UTC window, excluding gpt-5-mini and gpt-5-3-mini records while retaining blank model labels. Citation IDs are expanded from eligible answers. Parse and Soar source domains and subdomains are excluded. Latest content snapshots are selected strictly before October 7. SQL, raw output, and number trace are committed under docs/research-receipts for this slug. A pair is an answer, a distinct cited source, and a detected brand in its selected snapshot. Named means the brand ID appears in the answer brand array. Repeated answers count separately. Title and URL patterns classify listicle shape.
What this cannot tell us
- Only pages with measured brand content enter; snapshot content may postdate answers.
- No link extraction or manual validation of every prominence label was performed.
- Named does not necessarily mean favorably recommended.
- The tracked question panel is not a random sample of all AI search.
- This observational analysis does not establish a causal effect of a Signals service.
