How strict is the 95%level for species delineation for average nucleotide identity?

Question

[olp123]

Hello, I have 2 "training genomes" which have been already trimmed and quality checked (sequenced using Illumina HiSeq and MiSeq), and which should come from the same bacterial species. However, the JSpecies web server ist telling me that they have an ANIb value of 93.3% (Tetra calue is 0.99696) According to the literature, 95% is the threshold for ANIb. So, how strict are these 95%. Someone told me about 93%, but I could not find this value in the literature. Might the ANIb value change to >95% when I do not use the web server, but download the JSpecies Software and run again using different parameters?

What could have happened to my results? What can I do, to make them more similar and eliminate the mistake?

Thanks!