FE college recruitment in 2022-2025

21st May 2026 by Timo Hannay [link]

Just over a year ago, we were lucky enough to collaborate with the Gatsby Charitable Foundation and the Association of Colleges (AoC) to analyse a large collection of further educations (FE) recruitment adverts from AoC Jobs. We used gpt-4o, widely regarded at that time as the world's most capable large language model (LLM), to extract the kind of detailed information that had previously eluded our more traditional algorithms. The results covered adverts posted from May 2022 onwards and were published on this blog in November 2024. This post brings that analysis up to date with more recent data (to December 2025), improved analysis methods and a newer LLM model.

In summary, we find that:

• The majority of adverts (53%) were for support roles, with 45% for teaching roles and small minority for leadership positions.
• Nationally, construction and engineering were the most commonly sought subject groups for teaching positions, but there was significant regional variation. Maths and English were the most sought-after individual subjects.
• The vast majority of vacancies were for permanent positions, and there was no obvious long-term trend, with fixed-term and temporary positions accounting for around 10% of positions over most of the period studied.
• Similarly, part-time positions have been more or less stable at about 16%, though with considerable variation by type of role, being most common for teaching vacancies.
• In general, teaching and leadership positions offered more generous leave entitlements than support positions. Across all positions, pensions, training, parking and leave entitlements were the most commonly mentioned benefits. Gyms, pensions and parking have seen the biggest increase in mentions over the last two years.
• References to flexible working have been more or less consistent in recent years, at about 40% of adverts. They were more frequent for suport positions than for teaching or leadership positions. In most cases they appeared to be offered as a benefit rather than imposed as a requirement at the discretion of the employer, though there were exceptions to this generalisation.
• Salaries were lowest for support positions, somewhat higher for teaching positions and highest of all (with wide variability) for leadership positions. There was some regional variation too. Median salaries offered have been rising, but at below the rate of inflation.
• The most commonly required qualifications for teaching positions are (appropriately enough) teaching qualifications, degrees and Level 3 qualifications.
• We continue to see evidence of considerable repeat advertising, suggesting that many positions are not being filled first time around. Repeat advertising rates also appear to have risen, from about 25% in 2022 to around 35% in 2025.

 

Adding the ads

We analysed a total of nearly 171,000 adverts across 234 employers posted between May 2022 and December 2025. There was a gap in coverage during late 2024 and early 2025 due to the AoC Jobs website becoming temporarily unavailable to our crawler. Of these adverts, 48,681 came from one of 155 identifiable FE employers (corresponding to over 70% of the total number of FE colleges¹). This compares to 36,603 adverts from 134 FE employers used in our previous analysis.

Because we processed the content with a new LLM model (gpt-5.2), along with different prompts to those used in our previous analysis, we reanalysed the whole archive and did not rely directly on any results obtained last time². As before, we continued to use spot checks to confirm that the LLM was interpreting adverts consistently with what we would expect from a human reader, which in almost all cases it did. Overall, we observed similar patterns, which is reassuring, but in some cases our revised methods gave somewhat different results or revealed new trends; we comment on these below where they arise.

Figure 1 shows the numbers of adverts from FE colleges posted each month over the period studied. The seasonal patterns seen across all roles were similar for both teaching and support positions, with the largest peaks in July and smaller peaks in January, but also with considerable year-to-year variation. The ebbs and flows of leadership postions were much more random, in large part because there were far fewer of them. Note that we used a much tighter defintion of 'leadership' compared to our previous analysis, including only genuine senior positions at an institutional level and excluding miscellaneous heads of function. For the present analysis, we were able to more properly categorise that latter as support (or occasionally teaching) roles. (Show all roles again.)

(Use the menu below to select a role type to view. Hover over the lines to see corresponding data values.)

Figure 2 shows more clearly the breakdown by role type, with support roles representing 53% of adverts (compared to 45% last time) and teaching roles at 45% (down from 50%). Leadership roles accounted for only 0.8% of adverts (compared to 5% last time). A further 0.6% of adverts did not fall cleanly into any of these categories, which is why the proportions sum to just under 100%.

(Hover over the bars to see corresponding data values.)

 

Subject matters

As in our previous analysis, we examined teaching positions using a bespoke subject taxonomy³. Figure 3 shows the frequency of occurrence of the taxonomy's 30 top-level subject groups. You can also examine these by region. For example, 'Construction' comes highest in the North East and East of England, but lower in London and the South West. 'Languages, literature and culture of the British Isles' ranks higher in London and the South East than in the East of England and West Midlands.

(Use the menu below to select a region to view. Hover over the bars to see corresponding data values.)

Figure 4 provides a finer breakdown into subjects, but only shows the top 24 (of 104) from our taxonomy. As before, maths comes top, with English, Health, Engineering, and Computing completing the top five.

(Hover over the bars to see corresponding data values.)

 

So long?

Figure 5 shows the proportions of adverts that were for permanent positions versus various kinds of temporary ones. Across all roles, the vast majority were permanent, though support positions were more likely to be fixed-term or temporary than teaching positions or (especially) leadership positions. Compared to our previous analysis, temporary positions were more prevalent and those categorised as fixed-term or maternity appeared less frequently. This is at least in part due to variations in the LLM model and prompting. Where results differ, the analysis presented here should be considered more reliable.

(Use the menu below to select a type of role to view. Hover over the bars to see corresponding data values.)

Figure 6 shows how the proportions of fixed-term and temporary positions across all roles varied from month to month. Though there is considerable fluctuation between about 8% and 14%, there is no obvious seasonality or a clear long-term trend.

(Hover over the line to see corresponding data values.)

 

About time

Figure 7 shows the stated time commitment. 'None' indicates that this was not specified in the advert, which can often (but not always) be interpreted to mean full time. There is considerable variation between teaching, support and leadership roles. (Show all roles again.)

(Use the menu below to select a type of role to view. Hover over the bars to see corresponding data values.)

Figure 8 shows the proportions of part-time positions across all role types. As for temporary positions in Figure 6, there was considerable month-to-month fluctuation, with no obvious long-term term trend. However, there is a hint of seasonality, with peaks often occurring in the summer.

(Hover over the line to see corresponding data values.)

A minority of adverts mention the level of commitment in terms of a full-time equivalent (FTE) number between zero and one. The distributions of these are shown in Figure 9. Consistent with the results above, teaching positions were more likely than support positions to be around half-time. Leadership positions were too few to analyse separately. (Show all roles again.)

(Use the menu below to select a type of role to view. Hover over the columns to see corresponding data values.)

 

Fringe benefits

Figure 10 shows the distribution of leave entitlements mentioned in the adverts analysed. For part-time positions, we instructed the LLM to convert these into FTE values where possible. Instructions for ambiguous situations were stricter than in our previous analysis, resulting in a large reduction in retrieval rates (from 34% of adverts to just 10%). Yet a few very low values in the output suggest that in some cases a part-time leave entitlement was interpreted as a full-time one. (There is a similar issue with the salary data, as we will see below.) With that proviso, leave entitlements appeared to be somewhat more generous for teaching and leadership roles than for support roles.

(Use the menu below to select a type of role to view. Hover over the columns to see corresponding data values.)

In a new analysis this time around, we also extracted mentions of specific benefits on offer. The results for all FE college adverts are shown in Figure 11. Across the whole period covered, pensions (which appeared in nearly two-thirds of adverts), training, parking, leave and wellbeing time were the most commonly mentioned.

(Hover over the bars to see corresponding data values.)

In order to give a sense of how mentions of benefits have changed over time, Figure 12 shows the percentage-point changes between adverts posted in 2022-2023 and those posted in 2024-2025. Note that no attempt has been made to control for changes in (for example) employer or role composition, so these differences might have emerged for reasons other than general changes in preference or perceived utility, but we hope that they are nevertheless informative. Gyms, pensions and parking appear to have been the biggest risers.

(Hover over the bars to see corresponding data values.)

 

Flexibile friends

Figure 13 shows the proportions of adverts for all positions that mentioned flexible working. This increased between 2022 and 2023, but has since been more or less flat. Absolute levels are a bit higher than in our previous analysis (~40% versus ~35%), but that looked only at teaching positions, which, as we shall see below, tend to be lower than for support positions.

(Hover over the columns to see corresponding data values.)

Figure 14 shows how mentions of flexible working vary by type of role. They are highest for support roles, lowest of leadership roles and somewhere in between for teaching roles.

(Hover over the columns to see corresponding data values.)

Eagle-eyed readers will have noticed that overall mentions of flexible working shown immediately above in Figures 13 and 14 (~40% overall) are much more frequent than mentions in the context of the job benefits shown earlier in Figure 11 (~20%). What gives? In reality, flexible working is sometimes a requirement imposed on the employee at the discretion of the employer. This distinction is shown in Figure 15 (which is otherwise identical to Figure 14). About half of all mentions of flexible working are presented unambiguously as a benefit (blue segments). But in a much smaller proportion of cases it is clearly a requirement (red). In the remaining adverts, the distinction is not made explicit one way or the other (grey) – though our naturally generous tendencies incline us to believe that most of these are potential employee benefits rather than impositions.

(Hover over the columns to see corresponding data values.)

 

Down payment?

For adverts that mentioned a salary, we instructed the LLM to summarise this information, converting to an FTE basis for part-time roles where possible. When a range was provided, we took the middle value. We also filtered out any extremely low figures that were obviously hourly, daily or weekly rates rather than annual ones. The results are shown in Figure 16. As for the leave entitlements shown in Figure 10, some of the lowest value are improbable as FTE salaries (not least because they would be below the minimum wage for a full-time role), and appear to represent actual salaries for less-than-full-time roles. However, these complications shouldn't distract from the fact that the overall distributions look plausible, especially when comparing different type of role and – as we shall see below – changes over time. Generally speaking, support positions tended to be in the £20-30k range, teaching postions in the £30-40k range and leadership positions in the £60-70k range, with a much wider spread.

(Use the menu below to select a type of role to view. Hover over the columns to see corresponding data values.)

Figure 17 shows how the median salaries on offer have changed over time. We have not made any allowances here for the possibility that the compositions of job types and employers might have changed too, so we are not necessarily comparing like with like and the results should be considered indicative rather than definitive. But they are also based on large sample sizes, so should at least be informative.

Between 2022 and 2025, median teaching position salaries rose by ~15% and support positions went up by ~19%. But consumer prices rose by about 24% over the same period, which means that real wages on offer appeared to fall. All roles include leadership roles, but there were too few of those to split then out separately.

(Use the menu below to select a type of role to view. Hover over the columns to see corresponding data values.)

Figure 18 shows the same date by region. It is importat to note that sample sizes in the North East were small, so those results could be misleading. Across all roles, London had the highest median salaries on offer, followed by the West Midlands, North West and South East. See also the separate results for teaching and support positions. There were too few leadership positions to analyses in this way. (Show all roles again.)

(Use the menu below to select a type of role to view. Hover over the columns to see corresponding data values.)

 

Qualified offers

Figure 19 shows the top 10 qualifications required by applicants for teaching positions, where these were mentioned. Teaching qualifications, degrees and Level 3 qualifications (A-levels, T-levels, BTECs etc.) were the most common.

(Hover over the bars to see corresponding data values.)

 

On repeat

This section looks at putative repeat advertising in order to try and determine how often vacancies are not filled first time around and need to be re-advertised. The first step is to determine which adverts are similar enough to each other to be considered likely duplicates. To do this we use 'embeddings', which are numerical representations (technically, high-dimensional vectors) of the semantic content of a piece of text. We determined these for each advert using OpenAI's text-embedding-3-large model, which returns a string of 3,072 numbers in response to each piece of text it receives. Note that this is different from the text-embedding-ada-002 embeddings model that we used in our previous analysis, which uses only 1,536 'dimensions'. We would therefore expect the newer model to be more discerning. In order to determine the semantic similarity of any pair of adverts, we only need to compare their embeddings. Technically speaking, this involves calculating their 'dot product', which yields a number between 0 (no semantic similarity at all) and 1 (semantically identical).

Figure 20 shows the distribution of similarity scores (as percentages) for each pair of adverts in the data set that came from the same employer. The peak is at 77.5%, which is lower than in our previous analysis (around 86.5%) and consistent with the increased discernment of the newer model. We can also see that the tail to the right is longer, thicker tail than the one to the left, indicating a subset of advert pairs that are very similar to each other, including some that are virtually identical (ie, close to 100%). Last time we chose to define putative repeat adverts as those with similarity scores of 98% or higher. The greater sensitivity of the new model means that we could probably justify applying a slightly lower threshold, but since it is rather arbitrary anyway, we will continue to use the same cutoff value as before.

(Hover over the columns to see corresponding data values.)

Figure 21 shows the distribution of the number of days between posting dates for pairs of putative repeat adverts (ie, those with similarity scores greater than or equal to 98%). Note that similar FE colleges advertise certain positions (teaching assistants are a common example) almost continually. In many cases these probably aren't genuine repeats, so in order to prevent them from potentially skewing the results we filtered out any group of highly similar adverts with 20 or more putative repeats across the whole data set from May 2022 to December 2025.

Despite these changes to our methods, the results are very similar to last time. There are large peaks within a few days of the original advert. These seem to be immediate corrections or updates rather than true repeats. Then there are spikes at around one month after the original posting (presumably appearing as twin peaks due to the vagaries of weekends) and a clear weekly cycle. This suggests that it is common to wait a month or a whole number of weeks before reposting. As we pointed out last time, the existence of these regular patterns helps to confirm that at least some significant proportion of these similar-look adverts are real repeats.

Putting all this together, we define repeat adverts as those with >=98% similarity scores and appearing within 14-90 days of the orginal advert. Figure 22 shows the proportions of adverts that were followed by repeats. Across all roles, this is quite high and has been rising: from around 25% in 2022 to more like 35% in 2025, though with some month-to-month variation. Of course, the repeat rate falls precipitously towards the end of the time series because there has been insufficient time for repeat adverts to appear. (Last time we saw a higher repeat rate and no similar upwards trend. These differences are due to a combination of the new embeddings model and improvements to the underlying analysis, which now more accurately accounts for chains of three or more repeating adverts.)

Teaching and support positions showed similar trends, though repeat rates for the former tended to be slightly higher than those for the latter. (Adverts for leadership positions were too few to enable a meaningful analysis of those.)

Overall, there are certainly causes for concern – sub-inflation salary increases along with high and rising repeat advertising rates to name two. But we hope that this analysis has shone new light onto FE hiring activity over recent years, from subjects and qualifications to flexible working and other benefits. Once again, our special thanks to the Association of Colleges for allowing us to use the rich seam of information provided by AoC Jobs, and to the Gatsby Charitable Foundation for their domain expertise and financial support.

We're always pleased to hear from our readers. If you have comments, suggestions or questions, please write to us at: [email protected].

 

Footnotes:

1. This actually understates the coverage because some FE colleges are organised into federations, so a single employer in the AoC Jobs database can correspond to more than one FE college.
2. A further change concerns the exact source of some of the content we analysed. In our previous study, the job descriptions and employer information came primarily from the RSS feeds on the AoC Jobs website. Those have since been discontinued, so for this analysis all of the content came from the HTML pages, which in some cases contain somewhat different descriptive text. However, since we reprocessed the whole archive, all content represented here has been treated in a consistent manner.
3. To elaborate on the process, the job title for each advert was extracted by the LLM, and the results categorised with fuzzy matchig on keywords using the same custom taxonomies of roles and subjects described in our previous report.