How accurate are third party website traffic estimates, and what should a plan use instead?
Part of Fair publisher planning
The short answer
Third party traffic estimates are rough. A 2022 peer reviewed study of 86 websites found a commercial estimation service reported 19.4 percent fewer total visits than the sites' own analytics, with larger gaps for smaller sites. Rankings agreed better than totals. Plans should treat estimates as ranges, name the source and period, and prefer measured public data.
Key takeaways
- In a study of 86 websites, one estimation service underestimated total visits for 66 sites and came within about 5 percent of the sites' own analytics for only 5.
- Estimates built from panels or network data get weaker as traffic falls, because a small site leaves too few traces in the sample.
- A plan that cuts sites below a traffic estimate removes small and niche publishers on the weakest numbers it has.
- Coarse but measured public data, such as Chrome UX Report rank buckets, and publisher reported figures with stated definitions are stronger evidence than a single estimated total.
- Each traffic figure in a plan should carry its source, period, unit and a note on how certain it is.
How accurate are third party website traffic estimates?
The best public evidence comes from a peer reviewed comparison. Bernard J. Jansen, Soon-gyo Jung and Joni Salminen published a study in PLOS ONE (opens in a new tab) on May 27, 2022 that set 12 months of data for 86 websites from a commercial panel and network based estimation service beside the same sites' own Google Analytics numbers. They found "statistically significant differences between the two services" for total visits, unique visitors, bounce rates and session duration.
| Measure | Finding | What it means for a planner |
|---|---|---|
| Total visits, all sites | Estimated 19.4 percent lower in aggregate | Estimated totals can sit well below what the publisher records |
| Unique visitors | Estimated 38.7 percent lower on average | Audience counts diverge more than visit counts |
| Direction per site | 66 sites (76.7 percent) underestimated, 15 (17.4 percent) overestimated | The error has a direction, so it can tilt a ranking |
| Close agreement | 5 sites (5.8 percent) within about 5 percent | A close match for one site is the exception in this sample |
| Rank order | Rankings "significantly correlated, especially for total visits and unique visitors" | Estimates sort sites better than they size them |
Why are traffic estimates worse for small and niche sites?
Definition
- Panel based traffic estimate
- A traffic figure projected from the browsing of a sample of people or devices, such as users who installed a browser extension, scaled up to an estimate for the whole internet population.
Projection needs enough observations. The PLOS ONE authors note that "It is not clear that panel data would be accurate for lower-traffic websites as there is not enough traffic to these sites to generate reasonable statistical analysis," and that the difference between the two services "increased for the smaller websites." A regional food blog or a specialist trade site might be visited by a few panel members a month, and the estimate swings with each one.
Panels also miss people unevenly. The Media Rating Council's Digital Audience-Based Measurement Standards (opens in a new tab) (December 2017, version 1.0) define non-response bias as "the lack of or under-representation of a specific segment of the measured audience or the sample which can lead to erroneous research conclusions." If the readers of a niche site, such as a language community or a hobby group, are thin in a panel, the site's estimate inherits that gap.
How do traffic estimates bias media plans?
Bias enters at the threshold. A planner who keeps only sites above a monthly visit estimate is applying the weakest numbers to the smallest sites, and if the estimate leans low, as it did for 66 of the 86 sites in the study, the cut removes publishers that would pass on their own analytics. The publishers lost are independent, local and niche sites, the ones the fair publisher planning guide argues deserve an equal hearing.
The second bias is false precision. A spreadsheet that shows 41,200 monthly visits for one site and 39,800 for another invites a ranking the data cannot support. The CrUX metrics methodology (opens in a new tab) takes the opposite approach: its popularity rank is "a relative measure of site popularity within the CrUX dataset," placed on a log10 scale with half steps, and Chrome calls it "a guide for broad analysis."
What public data is more reliable than a traffic estimate?
No single source is complete, and the PLOS ONE authors describe "a triangulation of services" as the practical answer. What changes is how much weight each source deserves and what it actually measures.
| Evidence | Strengths | Limits |
|---|---|---|
| Chrome UX Report rank | Measured from real Chrome navigations; public; by country in BigQuery; dated monthly | Buckets, not counts; no record below an undisclosed threshold; excludes Chrome on iOS and Android WebView |
| Publisher analytics shared directly | Counts the site's own visits; can be cut by country, section and month | Definitions vary by tool; private unless the publisher shares it; not audited by default |
| Publisher reported totals | Public and dated; can be the only figure a small outlet has | Self reported; units and geography may be unclear |
| Accredited audience measurement | Methods audited against MRC standards; standard errors disclosed | Covers only the properties a measurement service includes |
| Third party estimates | Cover sites with no other data; useful to sort a long list | Projected, not measured; weakest for small and niche sites |
Read publisher figures for their units. Historica Canada's 2024/25 reach report (opens in a new tab) says The Canadian Encyclopedia "had more than 156,435,000 pageviews, from more than 14 million users." Pageviews and users are different units, for a full year, across all countries, and self reported. In Google Analytics, "Total users" is "The number of unique users who triggered any event in the specified date range," per the GA4 user metrics help (opens in a new tab).
How should a plan show uncertain traffic?
Borrow the disclosure rules that audience measurement already follows. The MRC's Minimum Standards for Media Rating Research (opens in a new tab) require each report to state that its data "are subject to both sampling and non-sampling errors" and to "contain standard error data relevant to the audience estimates." The IAB and MRC Audience Reach Measurement Guidelines (opens in a new tab), updated August 2013, add that "Known weaknesses in the projection processes should be disclosed to users."
Five rules for traffic figures in a media plan
- Name the source and period. Write which dataset or report the figure comes from and which month or year it covers.
- State the unit. Visits, pageviews, users, browsers and navigations are not interchangeable, so never put two units in one column.
- Show a range or a bucket. Round estimates to a band, such as a CrUX rank step, rather than a five digit number.
- Mark the certainty. Label each figure measured, publisher reported or estimated, and leave the cell blank with a reason when nothing exists.
- Keep traffic out of the ranking. Sort by brief fit and quality, and show traffic beside the result for pacing and budget decisions.
Traffic column review before a plan goes to the client
- Each figure names its source and period.
- Units match within each column.
- Estimated figures are labeled as estimates and rounded.
- Sites without data show a blank with a reason, not a zero.
- No site was removed on a traffic estimate alone.
- Figures from different sources are not added together.
How does PlacementGraph treat traffic data in a site list?
What does a traffic review change on a real plan?
Fictional example
A fictional ukulele maker in North Carolina and Virginia
An unnamed ukulele and small acoustic instrument maker asks its agency for a display plan in North Carolina and Virginia, beside music lessons, folk and bluegrass coverage and instrument care. The draft list applies a cutoff of 50,000 estimated monthly visits from a third party tool, which leaves six large music and entertainment sites.
The traffic review removes the cutoff and sorts the candidates by brief fit. Four regional music blogs and a luthier's publication fit the brief closely. Two have CrUX rank records in the top 1M bucket for the United States; three have none. One publisher shares a year of its own analytics, which shows roughly twice the visits the third party tool estimated, and the plan records both figures with their sources rather than choosing one.
The final plan keeps the six large sites and adds the five specialists, with a traffic column that reads measured bucket, publisher reported, estimated or no data, each with a date. Budget for the specialists is set as a test line, to be paced against delivery.
- Draft: six sites above an estimated traffic cutoff.
- Review: cutoff removed, eleven sites sorted by fit.
- Plan: eleven sites, each traffic figure labeled by source, unit and certainty.
Frequently asked questions
Are traffic estimates more reliable for large websites?
The PLOS ONE authors suggest panel data might be more accurate for larger sites, and their gaps widened as sites got smaller. Even so, aggregate visits across all 86 sites were 19.4 percent lower than the sites' own analytics. Treat a large site's estimate as a better sorted guess, not a count.
Can a planner ask a publisher for its own analytics?
Yes. Ask for a recent month or quarter, the tool used, the unit (visits, pageviews or users) and a country split if the tool records one. Record it as publisher reported in the plan, with the date, so the client can see it is first party data and not an independent audit.
Why do two traffic tools give the same site different numbers?
They measure different things in different ways. A site's analytics counts tagged visits by browser identifier; a panel tool projects from a sample of people; a browser dataset counts navigations from its own users. Each misses something different, so the same site can show different totals in each.
Should traffic estimates be removed from plans entirely?
No. They help sort a long list and flag sites that may be too small to carry a flight. The problem is using them as a hard cutoff or as precise numbers. Keep them as labeled estimates, rounded, beside stronger evidence where it exists.
Sources
- Measuring user interactions with websites: A comparison of two industry standard analytics approaches using data of 86 websites (opens in a new tab)PLOS ONE, Jansen, Jung and Salminen, May 27, 2022
- MRC Digital Audience-Based Measurement Standards, version 1.0 (opens in a new tab)Media Rating Council and IAB, December 2017
- Minimum Standards for Media Rating Research (opens in a new tab)Media Rating Council, revised December 2011
- Audience Reach Measurement Guidelines, version 2.0 (opens in a new tab)IAB and Media Rating Council, updated August 2013
- CrUX metrics methodology (opens in a new tab)Google Chrome for Developers
- [GA4] Understand user metrics (opens in a new tab)Google Analytics Help
- 2024/25 Reach Report (opens in a new tab)Historica Canada
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.