This is the feedback tool template overlap as a file. Seventy eight rows, one per vendor and page cluster, each carrying how many of that vendor's comparison or alternatives pages were fetched, how many were sampled, and the median and the largest vocabulary overlap between any two of them.
Overlap here is the Jaccard index of the sets of distinct words each page uses. A pair at 94 means the two pages share 94 percent of the vocabulary either of them contains. A pair at 99.8 means two pages of two and a half thousand words differ in about five words.
domain,cluster,pages_fetched,pages_sampled,median_overlap_pct,max_overlap_pct
featureos.com,compare,80,40,93.8,97.8
hellonext.co,compare,150,40,93.8,97.8
convas.io,compare,4,4,90.7,93.7
olvy.co,alternatives,20,20,89.2,99.8
Take the file. It may be copied, quoted and used commercially with no attribution and no link back.
None of it needs attribution. If you would rather point at it than copy it, the address is https://votefirst.app/en/data/feedback-tool-template-overlap and the figures on it were read on 5 October 2026.
What is template overlap, and why measure it?
It is one number for how much of a page was written and how much was filled in.
A vendor that publishes four hundred comparison pages did not write four hundred pages. It wrote one and substituted a name.
That is visible to a reader after two pages and invisible in any count of how many pages a competitor has, so a comparison of inventory sizes without this number compares four hundred pages to four hundred copies.
The measure is deliberately crude. It reads vocabulary rather than meaning, so two genuinely different pages about the same narrow subject will score higher than two pages about different subjects, and a low score is not a compliment.
What it catches reliably is the top of the range, where a pair sharing more than nine words in ten is a template by any definition.
Which feedback tool pages are the most templated?
FeatureOS and Hellonext, at a median of 93.8 percent across forty sampled comparison pages each.
Those two domains serve the same pages, so that is one template counted twice.
Convas follows at 90.7 percent across its four comparison pages, and Olvy's alternatives cluster reaches a median of 89.2 with a worst pair at 99.8 percent, which is two pages of about 2,584 words that differ in the name in the title and almost nothing else.
Four of the seventy eight rows sit above seventy percent. Ten sit at fifty percent or more. Thirty nine, which is half, sit under thirty percent, so the category is not uniformly templated and naming the four is more useful than a sentence about everybody.
What does a templated page mean for whoever is reading it?
That the page ranking above you may be one page wearing four hundred hats.
A page whose vocabulary is 94 percent shared with its siblings answers the question it names in about six percent of its words.
The other 94 percent is the boilerplate wrapped around the two names in the title, which matters before anybody decides a term is unwinnable because a large vendor already covers it.
The measure says nothing about which page ranks. A template can rank and often does. It says how much of the page is an answer.
How was this built, and when does it change?
By fetching the pages once, keeping each page's set of distinct words, and comparing every pair.
rival-map.py fetch collected the pages on 5 October 2026 and stored each page's word set. template-dataset.py reads that file, groups pages by vendor and cluster, computes the Jaccard index of every pair in a sample of up to forty, and writes the median and the largest.
A cluster with fewer than two fetched pages has no pair and therefore no row, which is why there are seventy eight rows rather than one per vendor.
The sampled column matters and is in the file for that reason. A vendor with 150 fetched comparison pages is measured on forty of them, so its median is a sample median and the file says which number it came from.
The refresh rule is one fetch per quarter, and the next is due on 5 January 2027.
What changed since the September fetch?
The top of the range did not move. FeatureOS, Hellonext, Convas and Olvy hold the same four rows above seventy percent that they held on 13 September 2026, at the same medians.
The movement is lower down. AnnounceKit's comparison cluster grew from 8 fetched pages to 36 and its median fell from 41.4 percent to 23.6, so the new pages share less wording than the first eight did. Formbricks went from 2 comparison pages to 10 and its median rose from 21.5 to 43.2. Gleap's alternatives cluster grew from 55 pages to 61 and its median rose from 32.1 to 45.3.
IdeaLift's two rows and Pendo's alternatives row dropped out, IdeaLift because its site answered with a server error and Pendo because its alternatives cluster fell below two pages, and AnnounceKit and Enterpret gained an alternatives row each. That is why the file holds seventy eight rows where it held seventy nine.
Who is this dataset for?
Anybody about to decide that a term is taken because a large site already covers it.
If you are counting the category rather than reading it, every page of all sixty nine vendors is enumerated in a second file, every heading at least two of them write is counted in a third, what each vendor charges is a fourth, and which directories listing them pass ranking credit is a fifth.
14 days, a card at signup, then $9 or $39 a month.