Why dwell time is a poor measure of content quality
Dwell time measures how long someone stays on a page, not why they stayed or whether they found what they came for. On short articles it was a workable proxy: three minutes on 500 words probably meant reading.
On long pages that logic breaks down. Take two visitors who each spend six minutes on a 3,000-word service page. One reads carefully and then requests a quote. The other skims the introduction, scrolls the whole page twice looking for a price indication and closes the tab. In the dashboard, both sessions look identical.
The engagement rate in GA4 does not fix this. According to Google Analytics help documentation, a session counts as engaged as soon as it lasts longer than 10 seconds, includes a key event or has at least two page views. The searching visitor in the example counts as engaged. The metric separates bounces from visits, but not a good page from a confusing one.
Google also does not describe quality in units of time. Its guidance on helpful, people-first content asks, among other things, whether readers leave feeling they need to search again to get better information, and states that Google has no preferred word count. A long page that leaves the deciding question open fails exactly that test.
What Haddad's dwell time research shows
Omar Haddad's 2026 study treats long dwell time as an ambiguous signal that can reflect interest or uncertainty. Published in the Journal of Contemporary Studies in Science, Technology, and Applied Research, it examines attention and brand visibility in AI-driven e-commerce, based on 41.7 million exposure events across eight markets. The conclusion: long dwell time does not reliably indicate content quality or buyer satisfaction. It can indicate confusion.
The core finding, in the study's own words: "A product page with rich descriptive content but unclear delivery or return information often receives dwell without conversion. In contrast, pages with moderate descriptive richness but clear delivery and return fields show lower dwell but higher add-to-cart rates."
The study treats delivery clarity, return policy visibility and payment options as measurable page signals. Extensive descriptions do little on their own; complete content combined with clear operational information works best.
| Page attribute | Effect reported by Haddad (2026) |
|---|---|
| Clear delivery information | +3.9% qualified attention, +2.8% add-to-cart probability |
| Visible return policy | Significant contribution to qualified attention, especially in Jordan, Egypt and Saudi Arabia |
| Rich description, unclear delivery and returns | High dwell, low add-to-cart |
| Moderate description, clear delivery and returns | Lower dwell, higher add-to-cart |
| Structured content | More evaluative and less ambiguous attention in most categories |
The last row comes from the study's ninth robustness check, which splits dwell by type: "Structured content increases evaluative attention and reduces ambiguous attention in most categories. Human influencer routes increase both evaluative and ambiguous attention, especially when the influencer path sends consumers to products with incomplete attributes."
Haddad draws the conclusion for anyone who steers by dwell time: "For platform managers and researchers, this distinction is crucial because optimizing only dwell may reward ambiguous pages that hold consumers longer without improving decision quality."
The effect is strongest where decisions are complex. In e-commerce that means categories such as electronics accessories and technical products, where specifications decide the purchase. Traffic source matters too. In the study's attention decay hierarchy, median attention is 74 seconds for organic search traffic against 34 seconds for short-video traffic, because visitors with a clear search intent are more likely to be evaluating.
The finding transfers well beyond retail. Delivery time and returns are to e-commerce what cost, duration, scope and cancellation terms are to a service. When those details are missing, an interested visitor stays longer without getting any closer to a decision.
Evaluative vs ambiguous attention
Evaluative attention is dwell accompanied by targeted interactions; ambiguous attention is dwell without such interactions that ends in an exit. Haddad (2026) introduces the two terms as a taxonomy of attention types: "Dwell without diagnostic interaction and followed by exit is treated as ambiguous attention." Both can be measured in any analytics tool.
Diagnostic interactions are actions that show someone is preparing a decision: clicking the pricing section, opening delivery or timeline details, expanding a FAQ item, viewing the contact form, choosing a variant, checking warranty or service terms. A visitor who does this is evaluating. A visitor who only scrolls and leaves has usually not found what they were looking for.
| Signal | What it measures | What it does not show | Value for content quality |
|---|---|---|---|
| Average dwell time | Time on page | Whether that time was productive | Low |
| GA4 engagement rate | Share of sessions over 10 seconds, with a key event or two page views | Whether the visitor found what they needed | Low to medium |
| Scroll depth | How far the visitor scrolled | Whether they were reading or searching | Low |
| Diagnostic interactions | Clicks on pricing, FAQ, delivery, contact | The final conversion | High |
| Key event (enquiry, purchase) | The business outcome | Why other visitors did not convert | High, but late |
The simplest test for any page: what share of long sessions includes at least one diagnostic interaction? If that share is low, the high dwell time is a symptom rather than a success.
Why operational clarity beats extra content
Operational clarity means a page answers the questions a decision depends on: what does it cost, how long does it take, what is included, and what happens if it does not go to plan. Descriptive content such as methodology, team introductions or vision statements builds credibility. It does not make a decision possible.
A prospect who has read a long methodology section and still has no idea of the order of magnitude of the investment is no closer to deciding than before. They stay because the text is interesting, and leave because that one number is missing.
The difference often comes down to single sentences:
| Buyer question | Vague wording | Operationally clear wording |
|---|---|---|
| How long does it take? | "Results within a reasonable timeframe" | "The standard programme runs 90 days in three phases." |
| What does it cost? | "Custom pricing" | "Engagements start at a fixed monthly fee that we state after the intro call." |
| What is included? | "Full support" | "Per phase: audit, implementation plan, monthly report." |
| What if it does not work out? | "We always find a solution" | "Cancel monthly after the first phase." |
The examples on the right are patterns, not advice on specific terms. What matters is that every answer can be checked.
How rarely business pages are written this way shows up in our research into the AI visibility of 310 SME websites: of 354 pages with figures or claims, only 3 linked to an independent source, and 41% of the sites had no machine-readable date. Concrete claims without a source or a date are hard to place, for visitors and for search systems alike.
Content quality and AI search: the same properties count
AI search systems favour concrete, well-supported statements, which are the same passages that help human visitors decide. A vague page is weak for both audiences.
A comparative study by Santhosh Kumar Iyappan (2026) found that factual accuracy shows a Very Strong positive correlation with AI trust signal ratings, the highest confidence level in the study. The same study identifies topical authority as the strongest cross-paradigm signal, with a Very Strong correlation across SEO, AEO and GEO at once.
Kargaev (2026) points the same way: content quality and relevance persist as signals across search paradigms. The quality AI systems assess is not descriptive richness but the operational specificity and factual clarity that also produce evaluative attention in people. Haddad (2026) builds the same bridge from e-commerce: a content improvement from the 25th to the 75th percentile yields an 8.7% gain in AI-assisted inclusion, and that gain comes from the same structured fields that shift dwell from ambiguous to evaluative.
The commercial weight is considerable. According to Iyappan (2026), AI-referred traffic converts at 14.2%, against 2.8% for standard organic traffic. Where that gap comes from is covered in our analysis of zero-click search behaviour.
Google puts it more soberly in its guide to optimising for generative AI features in Search: AI Overviews and AI Mode run on the regular ranking and quality systems, content should offer a unique perspective rather than restate what already exists, and there is no need to break text into tiny pieces for AI.
Whether AI systems use behavioural data such as dwell time at all has not been shown publicly. The practical conclusion holds regardless: adding operational details, giving them a source and a date, and placing them high on the page improves it for buyers and raises the chance that individual passages get cited. How content types differ in citability is the subject of our piece on AI content optimization.
How to separate interest from confusion in GA4
To tell evaluative from ambiguous attention in GA4, combine session duration with custom interaction events defined per page type. The effort is manageable if you start with your five most important pages.
- Define diagnostic interactions. Pick three to five actions per page type that show real progress in evaluation. For a service page: pricing section click, FAQ expansion, case study click, contact form view. For a product page: delivery information, returns policy, variant selection, size chart.
- Set up the events. Record each action as its own GA4 event. Clicks on page sections and FAQ accordions can usually be tracked with Google Tag Manager triggers, without touching the site's code.
- Segment long sessions. In an exploration, build two segments: long sessions with at least one diagnostic interaction and long sessions without. What counts as long depends on the page; the top quarter by session duration is a pragmatic starting point.
- Compare conversion rates. If conversion is clearly higher in the segment with interactions, dwell without interaction is mostly searching behaviour. Pages dominated by that pattern are the first candidates for more operational clarity.
- Check AI answers. Put your buyers' questions to ChatGPT, Perplexity and Gemini regularly and note which pages get cited. Why one platform is not enough for this is explained in our article on AI search monitoring.
Restructuring B2B service pages
B2B service pages are especially prone to ambiguous dwell because they explain the method at length and often leave out scope, duration and investment. A prospect reads with interest, grows convinced the provider knows its craft, reaches the end and leaves, still unable to tell whether the offer fits their budget and timeline.
The fix is not less methodology but a different order:
- Top of the page: what the service is, who it is for and what outcome it delivers, in one concrete paragraph.
- Early on: approach, phases, duration and the order of magnitude of the investment.
- Middle: methodology, team, case evidence. Long reading time is welcome here, because the visitor already knows they are in the right place.
- End: a FAQ covering the questions that hold up a decision, not general topic questions.
To make those details machine-readable as well, the title, description and structured data need to match; more on that in our article on metadata SEO.
Common mistakes when judging content quality
The most common mistake is reporting average dwell time as a success metric and holding up high-dwell pages as examples. The reporting then rewards exactly the pages where visitors have to search longest.
Three other patterns come up often:
- Burying operational information. Visitors who find no hint of price, duration or scope on the first screen either scroll at length or leave straight away. Both outcomes are worse than a clear statement near the top, even if that statement lowers dwell time.
- Treating the FAQ as topical coverage. "What is GEO?" answers no purchase question. "What does the programme cost?" and "When will we see first results?" do.
- Confusing depth with length. A 5,000-word page with 500 concrete words has less substance than a 2,000-word page that is grounded throughout. Why topical depth depends on density rather than word count is covered in our article on topical authority.
Campaign periods distort the same signals, as our piece on seasonal SEO shows.
Frequently asked questions
Is long dwell time bad for SEO?
No, long dwell time is not bad in itself, but it is not proof of quality either. On an in-depth knowledge article it can mean thorough reading. On a service or product page without pricing or process information, it more often points to visitors looking for something that is missing. Haddad (2026) found that pattern in 41.7 million exposure events: rich pages with unclear delivery and return information received plenty of dwell and few add-to-carts.
Does Google use dwell time as a ranking factor?
Google has not publicly confirmed that it does. Its helpful content guidance describes quality through questions about the content, such as whether readers need to search again afterwards, not in units of time. For your own analysis, dwell time stays useful as long as you combine it with interactions.
What is a diagnostic interaction?
A diagnostic interaction is an action showing that a visitor is preparing a decision. Typical examples are clicks on pricing or delivery details, expanding a FAQ item, choosing a product variant and opening the contact form. You decide per page type which actions count.
Is the GA4 engagement rate enough as a quality metric?
No, because GA4 already counts any session longer than 10 seconds as engaged. The metric filters out immediate bounces but does not distinguish a visitor who is evaluating from one who is searching in vain. For that you need your own events for diagnostic interactions.
Does operational clarity make pages shorter or worse?
Usually not, because operational details add to content rather than replace it. A price indication, a duration and a scope description make a page only slightly longer, yet answer the questions decisions depend on. Average dwell time may fall afterwards while enquiries rise. Haddad does not argue that rich content is bad: rich content without operational specificity produces confusion dwell, and the combination of both produces the most evaluative attention.
Does the dwell-as-uncertainty pattern apply to all content equally?
No, it is strongest where decisions are complex. Haddad (2026) reports it most clearly in categories such as electronics accessories and technical products; on the web, the equivalent is any service with a significant cost or time commitment. Low-stakes decisions show less ambiguous dwell, because visitors either decide quickly or leave quickly. Traffic source adds to this: intent-driven organic visitors (74 seconds median attention in the study, against 34 for short video) are the most likely to evaluate.
Does operational clarity also help visibility in AI search?
Yes, concrete and well-supported details are easier for AI systems to cite than general promises. Google recommends no special techniques for its AI features, only helpful content with its own perspective. A page with clear figures, a source and a date meets both requirements.
Want to know which of your pages leave visitors searching instead of deciding? Request a free analysis of your website.
