Such claims ("resulting in the intensity of traffic staying the same") depend on a lot of things, the time frame and region considered most of all. A mini-review of prior work in the introductory part of a 2022 paper (I'm omitting the table of the studies) says:
The mentioned studies found vehicle distance travelled to lane length elasticities between 0.16 and 1.39, with most ratios are between 0.2 and 0.9 with various time periods, countries of origin and methodological considerations. High variation implies that induced traffic is indeed highly dependent on the circumstances, especially on existing network characteristics and the volume of latent demand.
Likewise a 2018 UK study/review:
The evidence reviewed in this study supports the findings of the SACTRA (1994) report that induced traffic
does exist, though its size and significance is likely to vary in different circumstances. [...]
Findings for state level road networks in the US and the national Dutch network indicate an elasticity of
around 0.2 across the whole road network, i.e. a 10% increase in road capacity could lead to 2% induced
demand on the network. [...] Induced demand is likely to be higher for capacity improvements in urban areas or on highly congested
routes.
That even mentions/summarizes the study in the Q:
Duranton & Turner (2011), use data from 228 Metropolitan Statistical Areas (MSAs) in the US and find large elasticities for IH demand with respect to IH capacity in the range 0.82 to 1.39, with the authors preferred value, 1.03, across all estimation methods used.
OTOH
Pasidis (2017) determines elasticities for 545 Large Urban Zones (LUZ) in the EU28. Elasticities for the study period are found to be in the range 0.7 to 1.0. [...] Pasidis (2017) finds a much smaller elasticity for urban zones with metro systems (0.2), than for those without (0.72).
The review also mentions that the vast majority of case studies have been conducted in the US, where no alternative transportation methods generally exist on the scale/region considered. And regarding the studies that reported near 1 elasticity:
As they focus on particular road types, the demand response reported in these studies generally include re-assignment effects and is larger than the induced demand response.
Whereas studies that try to account for re-assignment calculate induced demand differently:
Case study evidence for the UK comes from Sloman et al. (2017) - who looked at a range of road
improvements - and Rohr et al. (2012) who calculated induced traffic for the Manchester Motorway Box .
Sloman et al. (2017) use a screenline approach to control for re-assignment and also report that they control
for background growth. They calculate induced traffic as the percentage change in traffic flows, where traffic
flows are based on trips (AADT or equivalent) and not distance travelled. As noted earlier, these measures are
reported to be in excess of background traffic growth, which has been based on the average regional and
county comparators over the same period. They found induced traffic for eight out of nine schemes in the
range 5 to 10 per cent but 20 per cent for the M25. They report a short-run average increase of 7 per cent for
seven schemes and a long run average of 47 per cent based on six schemes for which data were available 8 to 20 years after implementation.
And even for the US, results appear to differ drastically based on methodology:
Hymel, Small and Van Dender (2010) used 1966-2004 U.S. state-level cross-sectional time series
data to evaluate how income, fuel price, road supply and traffic congestion affect vehicle miles
travel (VMT). They find the elasticity of VMT with respect to statewide road density is 0.019 in
the short run and 0.093 in the long run (a 10% increase in total lane-miles per square mile
increases state vehicle mileage by 0.19% in the short run and 0.93% in the long run); with
respect to total road miles is 0.037 in the short run and 0.186 in the long run (a 10% increase in
lane-miles causes state VMT to increase 0.37% in the short run and 1.86% over the long run);
[...] Their analysis indicates that long-run travel elasticities are typically 3.4–9.4 times short-run elasticities.
And that paper cites a meta-conclusion:
Induced travel effects generally decrease with the size of the unit of study – Larger effects are
measured for single facilities while smaller effects are measured for regions and subareas.
This is mainly due to diverted trips (drivers changing routes) causing more of the change on a
single facility, whereas, at the regional level, diverted trips between routes within the region
are not considered induced travel unless the trips become longer as a result.
And the time frame also matters, e.g. a 2022 MsC thesis on the US, which considers only more recent times:
I find that between 1980 and 2019, total lane miles increased by 13%, resulting in 8% to 24% more VMT. I also find that population growth results in 41% more VMT and rising per capita incomes result in 19% more VMT, driving most of the increase in vehicle miles traveled. Other factors
contribute 7%, with an important portion of the increase unexplained.
