How we count certificates in job ads

What we measure

We measure the share of a role's job ads that name a certificate. We take the ads for one job title in one country that contain the certificate's name and divide them by all ads for that job title in that country, in the same month.

This is a narrower number than the "job count" many sites publish. A raw count of every ad that mentions the PMP mixes project managers with consultants, recruiters and course sellers. Dividing by the ads for one role tells you how often employers hiring for that role bring the certificate up.

Two sites can query the same Adzuna data and publish very different counts for one certificate. One may search all ads, another only ads with a given title. One may search "PMP," another "Project Management Professional." Another may count on a different day. That is why every figure here comes with its query.

We count a small set of skills per role (for example SQL, Jira or Kubernetes) in the same way, so a page can set a certificate's share next to the skills the same ads name. We chose these skills ourselves. They do not come from an occupation database.

Definitions

The certificate count is a strict subset of the role total: both use the same title search, and the certificate query only adds the phrase. A share can never pass 100%.

Worked example. In October 2026, the US search for the title "network engineer" returned 9,475 ads. Of those, 1,557 contained the phrase ccna. The share is 1,557 divided by 9,475, times 100: 16%. The CCNA is counted in one role only, so on an English page that is not written for one country, this is its home-country share and sets its verdict label. The search-terms appendix below lists which roles have figures this month; a role marked "not counted yet" has none. Our catalog also assigns the PMP to program-manager ads, and until that search has run, every PMP figure comes from project-manager ads only.

What we do not measure

• Hiring outcomes. A share says how often ads name a certificate. It does not say whether holders get hired faster, get more interviews or keep their jobs longer.
• Requirement strength. Our query cannot separate "required" from "preferred" or "nice to have." We write that an ad names or mentions a certificate, never that it requires it.
• Salaries from Adzuna. Our permission from Adzuna covers ad counts, not salary data, so no pay figure on this site comes from Adzuna. Pay figures come from public statistics (the BLS in the US, the ONS in the UK) or from surveys we name, such as a certifying body's own salary survey.
• Jobs outside job ads. Roles filled through referrals, internal moves or recruiters who never post an ad are not in the data.
• Trends. Until we have several months of counts, we report what one month shows and make no claim about direction.

The exact query

Each month we send three kinds of query to the Adzuna search API for every country and job title we track. We read only the count field Adzuna returns (the number of matching ads); we do not store or republish the ads themselves.

For example, the US project-manager figures come from these two calls (keys removed):

https://api.adzuna.com/v1/api/jobs/us/search/1?title_only=project manager&results_per_page=1
https://api.adzuna.com/v1/api/jobs/us/search/1?title_only=project manager&what_phrase=pmp&results_per_page=1

The skill phrase "agile" appeared in 12% of those US project-manager ads.

How we choose a phrase. A certificate phrase is never a generic word or a job title. It is the certificate's full name or a code that means nothing else in a job ad. The PMP is searched as pmp; the CISA is searched by its full name (certified information systems auditor) because CISA is also the name of a US government agency. A phrase is marked approximate when it can miss or overcount ads for the certificate: it names a whole family (such as azure certification), has a "+" removed or uses one common wording of a name that ads write in several ways.

Certificates with a "+" in the name. Adzuna's search ignores the "+" character, so "CompTIA Security+" cannot be searched as written. We search comptia security instead and mark the figure as approximate. Ads that write only "Security+" without "CompTIA" are missed, so these shares are more likely to be too low than too high.

Local credentials. A certificate that exists in one market is counted only there: GPM only in Germany, APM PMQ only in the UK.

Checking a figure yourself. Next to every figure, "Source: Adzuna" links to Adzuna's public search for the same title and phrase. The public website searches the whole ad, not only the title, and its results change as ads come and go, so its count will not match ours exactly. The API calls above are the method; the link shows you the kind of ads behind the number.

Which titles and phrases we search

The complete list of search titles (per role and country) and phrases, with the approximate ones marked, is generated from our catalog and published under Search terms we use. It is built from the same catalog and settings files our collection script reads, so the list and the method cannot drift apart. Three entries from it, as an example:

Countries and job titles

We collect data for ten countries: the United States, the United Kingdom, India, Brazil, Mexico, Spain, France, Germany, Italy and Poland. If a country's counts for a month are not in yet, the page leaves those figures out.

Each role has one search title per country. In the US, UK and India we search the English title ("project manager," "data analyst"). In the other seven countries we search the title local ads use: "gerente de projetos" in Brazil, "chef de projet" in France, "projektmanager" in Germany, "kierownik projektu" in Poland. Where local ads keep the English title (for example "project manager" in Italy, or "data analyst" in France, Germany and Italy), we search the English one. Three roles (Scrum master, product owner and UX designer) use the English title in every country, because that is the title local ads use.

This choice has two limits, and both make the role totals in non-English countries smaller than the real market:

• One local title, not every variant. German ads use "Projektleiter" as well, French ads "responsable de projet." Ads with those titles are not in our totals.
• The English-title gap. In Germany, France, Poland and elsewhere, many tech employers write their ads in English. A Berlin ad titled "Project Manager" is not counted under "projektmanager." This matters most for software, data and cloud roles, where English ads are common.

A smaller role total makes each share less stable, which is why the thresholds below exist. When we compare countries on a page, we compare shares, not raw counts, because the totals are not built on the same title coverage.

Thresholds, small totals and rounding

A count is not a random sample. It is every matching ad on Adzuna on the day of the query. No margin of error applies in the survey sense. Small totals are unstable for another reason: a handful of ads coming or going moves the share. Our chart notes use the word "sample" in that plain sense.

• Under 50 ads for a role in a country: we do not publish a share. The page leaves the figure out or says there are not enough ads.
• 50 to 199 ads: the small-total flag is set. We publish the share with a mark and the note "Fewer than 200 ads in this sample: read as a rough signal." The same mark appears in tables and charts. Heat-map cells for these totals are left blank.
• 200 ads or more: the share is shown without a mark.

Rounding. The share is stored unrounded (to three decimal places) and rounded only when shown. The same rule applies to every chart, heat-map cell, table, tool, quiz and sentence on the site:

The illustrations are made-up values that show the rule, not data. A share just under 1% is always a count, never 1.0%. The 10% line applies to the rounded value, so 9.96 reads 10%. In heat-map cells and the tool's bars, where space is short, the count appears without its total ("20 ads").

Verdict labels. Certificate pages carry a label set by fixed thresholds on the home-country share. The label is "Employers ask for it" at 15% or more, "Sometimes asked for" from 3% to under 15% and "Rarely named" under 3%. It uses the unrounded share, so a share displayed as 15% can carry "Sometimes asked for" when the stored value is just under 15. If the home country has no published share, the page shows no label. Nobody picks a label by hand.

The thresholds are editorial cutoffs, not statistical tests. They are fixed in the site's code, and we will change them only with an entry in the changelog below.

Refresh cadence and limits

We run the count once per calendar month. One full run is about 2,400 queries (role totals, certificates and skills across ten countries). We send at most 24 queries a minute, 240 a day, 960 a week and 2,400 a month, just under Adzuna's default API limits, so a run takes about three weeks.

Queries run in order of page priority. Within each priority tier, a role's total is collected first, because shares depend on it, then the certificates for that role, then its skills. Two consequences follow:

• A role total and a certificate count for the same role can be taken on different days of the same month. Ads come and go daily, so a share can move a little for that reason alone. We store the date of every query. As of October 3, 2026, every certificate and skill count collected for October 2026 was taken on the same day as its role total.
• Each count is a snapshot of the ads live on Adzuna on the day of the query, not every ad posted during the month.

A query that fails is retried in a later pass. A missing count is left out, never estimated. A page always shows the latest month we have collected; if a figure for that month is not in yet, the page does not fall back to an older number.

Known biases

Every method misses something. These are the limits we know about, and which way each one pushes the numbers.

• Title matching. We count ads by title. "Project manager" includes senior, junior, construction and IT project managers alike, and a certificate that matters in one industry is diluted by the others. Ads for the same work under another title ("delivery lead," "service desk analyst") are not counted. An ad whose title fits two roles can be counted under both.
• Phrase matching. An ad that spells a certificate out ("Project Management Professional") without the short form is missed when we search the short form. We do not add the counts for the short form and the full name together, because an ad that uses both would be counted twice. A phrase counts any mention, including "PMP not required" or a training benefit ("we pay for your PMP"). We have not measured how often this happens.
• Generic skill words. Some skill phrases, such as "agile," are also everyday words ("an agile team"), so skill shares can run higher than real demand for the skill.
• Required vs preferred. As above, every mention counts the same.
• The "+" stripping. Shares for CompTIA A+, Network+, Security+, Linux+, PenTest+ and Cloud+ are approximate and probably low.
• Duplicates. Adzuna collects ads from many job boards and employer sites, and the same job can appear more than once (reposted, or posted by both the employer and an agency). We use Adzuna's count as it comes and do not remove duplicates ourselves. Duplicates inflate the role total and the certificate count alike, so they distort shares less than raw counts.
• Aggregator coverage. Adzuna does not carry every ad from every employer site or job board, and its coverage differs by country. Coverage is broadest where the totals are largest (the US); small totals, such as some roles in Mexico or Spain, swing more from month to month.
• Ad text as indexed. We search the text Adzuna holds for each ad. If an ad reaches Adzuna shortened, a certificate named in the missing part is not counted.
• Language. Ads in a second language (an English ad in Poland, a Spanish ad in the US) are counted only if their title matches the title we search for that country.

How we verify certificate facts

Exam fees, formats, prerequisites, renewal rules and similar facts live in one fact file per certificate, shared by every page and every language. Writers use these values and never re-research them on a page; only the fact-check stage changes them.

Each fact carries a source link, a verbatim quote from that source and a confidence level:

On October 3, 2026, we re-read the source of every stored quote in our 105 fact files. Of 1,407 facts, 1,133 quotes matched their source word for word and 149 were replaced with the exact source text. The other 125 were not part of that check; 25 of them are prices read in a browser on the vendor's site on the same day. Each fact file records the date and result of the check. This was a one-off audit, not a recurring job; we will date the next one here.

Fees and other facts that vendors show only inside a store or a script are read in a browser on the vendor's site, and the date is recorded. Where a vendor uses regional prices, we say so and give the source for each region. Currency conversions show the rate and its date.

Charts are generated from the data and the fact files, and their captions are computed from the same values, so a caption cannot state a number the chart does not show. A lint check runs on every build and fails it on unformatted values, empty or broken labels and a missing source mark.

How pages are reviewed

Every page goes through the same stages before it can be indexed by search engines:

1. Research and draft. A research file lists every claim with its source, a quote and the date checked. Our own numbers enter the draft only as placeholders that the site fills from the data, so no job-ad number is typed by hand.
2. Fact check. A separate reviewer checks every claim against the research file and the sources, and removes or corrects anything unsupported.
3. SEO check. A script checks the page structure and links, and checks that the page does not repeat other pages.
4. Critic panel. Three reviewers read the page as a reader from the page's country, as a specialist in the field and as an SEO editor. Each must score it at least 9 out of 10; otherwise the page is revised and reviewed again.
5. Language check. A native-language reviewer reads the whole page for language only. A page is published only on a "publish" verdict; every error found is added to the writer's instructions so it is not repeated.

We use AI tools to draft pages and to run several of these review stages. That is why the rules are strict: every number comes from our data or a cited source, every fact has a stored quote and no page is opened to search engines until each stage has passed. Until then a page is visible but marked so that search engines do not index it.

Corrections policy

If you find an error, contact us through our About page, where we list how to reach us. Tell us the page, the sentence and, if you can, a source.

• We check every report against the source. If a fact is wrong, we fix it in the fact file, so the correction reaches every page and language that uses it.
• A factual correction is recorded in the update log of the page it affects, with the date and what changed. Corrections to the method are recorded in the changelog below.
• A new month of data is an update, not a correction. Changes in the numbers between months are not logged as errors.

Independence and how we make money

CertWorthIt is free to read. It earns money from affiliate links (for example to course platforms), and may show ads later. Affiliate links appear only after the verdict, with a disclosure, and they never change it: verdict labels come from the share thresholds above, and the text is written from the data and the sources before any link is placed. No certifying body, course platform or training provider can pay to be included, ranked or reviewed. Adzuna provides the data but does not see or approve our pages.

Who we are

CertWorthIt is a small independent site. Pages are edited by Elena Marsh, the pen name of the site's editor. Her role is the one described on this page: she runs the monthly data collection and checks every page against its sources before it is published. The byline makes no claim to certifications, employers or years of experience in the fields we cover; the pages rely on the cited sources instead. The publisher's details and contact information are on our About page.

How to cite our numbers

You may quote and cite any figure on CertWorthIt with a link to the page you took it from. Always give the share together with its country, job title and month; a share without them can be read wrongly. A suggested format:

CertWorthIt (October 2026). Share of US project-manager job ads naming the PMP. Based on Adzuna API counts. https://www.certworthit.com/certificates/pmp/

To cite the method itself:

CertWorthIt. How we count certificates in job ads, method version 1.0, last updated October 3, 2026. https://www.certworthit.com/methodology/

The shares are calculated by CertWorthIt. The ad counts behind them come from Adzuna, so please name Adzuna as the underlying source. We do not offer a download of the raw counts: they are Adzuna's data, and our permission covers showing them on our pages. Each figure links to its Adzuna search instead.

If you spot an error while citing us, use the corrections process above.

Data credits and licenses

• Adzuna. Job-ad counts from the Adzuna API, used with Adzuna's written permission (September 30, 2026). Every figure carries "Source: Adzuna" and a link to the matching search. Salary data and individual listings are not part of that permission and are not used.
• BLS and ONS. US pay and outlook figures come from the Bureau of Labor Statistics, with the access date; the BLS cannot vouch for the data or for any results derived from it. UK earnings come from the Office for National Statistics under the Open Government Licence.
• Reddit. Pages summarize what people report in public Reddit threads. We quote at most one or two short sentences from a post or comment, always with a link to it, and never show usernames. These posts come from people who chose to write about their experience, so they are not a representative sample; we present them as reported experience, not as data.
• Planned: O*NET and ESCO. We plan to add occupation and skill data from O*NET (US) and ESCO (the European classification of skills, competences and occupations). No page uses either today, and the skills we count are our own list. When a page does use them, O*NET data will be credited to the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA) under the CC BY 4.0 license. ESCO labels will be credited as © European Union, used under the Creative Commons Attribution 4.0 International (CC BY 4.0) license, and we will say where we changed a label. Parts of ESCO's skills data come from third parties under their own licenses, as listed in ESCO's copyright notice.
• Our shares. The shares and verdict labels are compiled by CertWorthIt. See How to cite our numbers.

Certificate and vendor names are used only to describe them. CertWorthIt is not affiliated with or endorsed by any certifying body.

Changelog

• October 3, 2026: Method version 1.0. Added exact definitions, rounding rules, the quote audit, a citation guide and the search-terms list.
• October 2026: First monthly count (October 2026 data). Methodology page drafted, describing the method as used for that count.

Search terms we use

These tables are generated from our catalog and pipeline settings on every build, so they always match the queries we send. Role totals use title_only with the job title; certificate and skill counts add what_phrase. Countries: United States, United Kingdom, India, Brazil, Mexico, Spain, France, Germany, Italy, Poland.