We investigate the discrepancy between oxygen abundance estimations for narrow-line regions (NLRs) of Active Galactic Nuclei (AGNs) type Seyfert 2 derived by using direct estimations of the electron temperature (Te-method) and those derived by using photoionization models. In view of this, observational emission-line ratios in the optical range (3000 < \lambda(\AA) < 7000) of Seyfert 2 nuclei compiled from the literature were reproduced by detailed photoionization models built with the Cloudy code. We find that the derived discrepancies are mainly due to the inappropriate use of the relations between temperatures of the low (t2) and high (t3) ionization gas zones derived for H II regions in AGN chemical abundance studies. Using a photoionization model grid, we derived a new expression for t2 as a function of t3 valid for Seyfert 2 nuclei. The use of this new expression in the AGN estimation of the O/H abundances based on Te-method produces O/H abundances slightly lower (about 0.2 dex) than those derived from detailed photoionization models. We also find that the new formalism for the Te-method reduces by about 0.4 dex the O/H discrepancies between the abundances obtained from strong emission-line calibrations and those derived from direct estimations.