There are many good answers with general advice already, instead, I will show you what the game rules advise on how you should be handling this since the core rulebook is vague about social conflicts in general.
Then, it describe many possible scenarios of characters using Bluff that could (and will) come up during play, be that on social conflicts or not:
This section includes clarifications and details on several different ways to use Bluff—and on several things that don’t work. (...)
Bluff Doesn’t Define a Response: Even the most successful lie told using Bluff doesn’t determine the course of action the deceived person takes—it just primes the target with misinformation.
This means attempts to trick a creature into a course of action might need to also include Diplomacy or Intimidate after the Bluff check. For example, suppose there was a guard with the following orders from the guard captain: “Don’t let anyone into the restricted area without clearance papers, even if it seems to be me or someone of higher rank.” After this, a sneaky rogue attempts the following ruse: “I am the king’s general on a mission of utmost importance for national security. I need you to let me in now, or you’re fired!” Assuming the rogue succeeds at her Bluff check, the guard now believes her to be the general, but this doesn’t mean he will let her through.
His orders still require him to keep everyone out without papers. The last part of the rogue’s demand is an attempt to Intimidate the guard, and the successful Bluff check was a necessary prerequisite to even attempt the Intimidate check.
Circumstances: When using Bluff to tell a lie, the Core Rulebook table on possible circumstance modifiers takes into account several levels of plausibility, targets who want to believe or are impaired, and possession of convincing proof, but there are also plenty of other circumstances that might affect the result of a Bluff check. For instance, many people strongly don’t want to believe a bluff that would lead to cognitive dissonances, such as attempting to convince a true believer that their religion is fake, and such a lie imposes a –5 penalty on the attempt (the opposite of a target who wants to believe the falsehood).
On the other hand, a target who is afraid that the deceit is actually correct might grant a +2 bonus or more on the skill check, depending on the level of anxiety about the fabrication. For instance, a bigoted assassin who is afraid that half-orcs are cannibals might be more likely to believe a half-orc’s bluff that she ate the target he was supposed to kill.
There are a variety of other circumstances, all of which might alter the odds in different directions. A character with a widespread reputation of being a compulsive liar might take a large penalty on his skill check, but a character with a reputation for always telling the truth, such as a paladin, would gain a large bonus on her skill check. Similarly, a hostile creature is much less likely to believe a deception, whereas a helpful creature is much more likely to believe one.
Tricking Someone: Bluff can be used to cleverly trip a target up and get him to reveal something or make a mistake. In these cases, he realizes his mistake soon after, but by then it is too late, and the falsehood has done its damage. This is similar to using Bluff to feint or create a distraction, but has broader applications in social situations.
For instance, suppose a swashbuckler suspected that an assassin works for the queen. The swashbuckler might be able to trick the assassin into revealing more information by pretending to be a fellow agent of the queen in an attempt to gauge the assassin’s response. Of course, if the assassin doesn’t work for the queen and sees through the ruse, he might attempt his own Bluff check to pretend that he works for the queen and fell for the trick, thus causing the swashbuckler to investigate the innocent queen.
Plausibility: The Core Rulebook mentions that some lies are implausible enough that no matter how high a character’s Bluff check, a PC can’t convince a target that they are true. However, the same page also presents a table that says that “impossible” lies impart a –20 penalty on the skill check. This table’s entry might actually be better described as “particularly implausible.” For example, an older human woman telling a very similar-looking human girl that she is herself from the future might take the –20 penalty, whereas a 10-year-old half-elf telling a 40-year-old orc the same lie would automatically fail the Bluff check.
You’re Not Lying, You’re Just Wrong: Sometimes a character is a convincing enough liar that targets can’t tell the character is lying, even when the targets possess incontrovertible proof that what the character is saying isn’t true, or the lie is otherwise too unbelievable to be possible. In this case, one way to resolve the situation is for the bluffing character to take a –20 penalty on the skill check, and if she beats the target’s Sense Motive, then the target believes that the bluffing character isn’t lying, but is simply mistaken.
This could also be the result of other situations in which the target of the Bluff attempt has strong reasons to believe that the falsehood, despite being plausible, isn’t factually correct. Even this result can be useful to the bluffing character, as it doesn’t mark her as a liar, and it allows her to gather information about what her target knows and expects.
True Lies and Implausible Truths: Bluff is the skill that convinces someone that something is true. However, there are a few potential cases when the situation isn’t as straightforward as a bluffing character telling a lie to a target.
For instance, suppose that the bluffing character makes up a believable lie to tell the target, and the lie turns out to be true, unbeknown to the bluffing character. If the Bluff check succeeded, the target is convinced, and might later verify the truth and trust the bluffing character more. However, what if the bluffing character fails? In this case, the target can tell that the bluffing character is lying, but that doesn’t necessarily mean that the target is forced to conclude that the information is false.
For instance, suppose a popular king has fallen into a magical, unbreakable sleep. A charlatan, noticing the king’s lack of public appearances, makes up a story about the king being placed under a sleeping curse and spreads it around the tavern, but his Bluff check is terrible, and everyone can tell he’s making it up. If one of the king’s advisors is present in the tavern, this doesn’t mean that the advisor now thinks the king isn’t in a coma; it just means that she can tell the charlatan doesn’t believe his own story.
The reverse side of true lies is implausible truths. These are situations in which someone is telling the truth (either saying something that is actually true, or spreading a lie that they believe to be true), but that truth is extremely implausible to the listener. Though the bluffing character isn’t lying, the same skill set that makes an excellent and convincing liar could potentially help characters attempting to spread an implausible truth.
In these cases, even if the target succeeds at the Sense Motive check, he can tell that the bluffing character truly believes what she is saying, and he might simply conclude that she isn’t lying, but simply mistaken. The target might later be swayed if presented with evidence or through a verbal duel (see pages 176–181). If a bluffing character successfully convinces a target of a lie and the target attempts to spread that information, this leads to a classic example of an implausible truth.
Personally, I never roll for Bluff when GMing NPC's (unless its being used in combat), I use 10 + Bluff bonus as the DC of the PC's Sense Motive check. Those checks may be asked on demand by the players, and are rolled in secret (even if the NPC is not actually bluffing). Also be affirmative on the result: "She is speaking the truth" and not "She appears to be speaking the truth". Unless your intent is exactly to get them on the edge and suspicious of everybody they meet (which is fine for a horror game).
When discovering this bluff may be important to the history and has roleplay potential, I will do the opposite and roll all players Sense Motives against the bluffer (also in secret), using 10 + Sense Motive bonus as the DC for that check.
If one of the PCs pass on that check, I will inform that the PC(s) notice that something is not right in what they are being told, but not exactly what, they will have to roleplay their way out of it and possibly make more Sense Motive checks, along with some Bluff and Diplomacy checks. The NPC may simply be hiding something, they may be afraid of being discovered, or are afraid of their boss, or are sad or angry about something they just mentioned, and so on. They have something to hide and that something could be discovered by the players, so a Bluff/Sense Motive check is warranted.
Best Answer
Yes.
Bluff has the rules for unbelievable lies, and also the GM is encouraged to alter DCs (or apply appropriate penalties/bonuses) to reflect strange or ridiculous circumstances. But keep in mind, the highest human bluff score in our world is like, +8*. And con artists can and have convinced people to leap off buildings, unlock doors they are guarding and hand the con artist the key and their gun, or betray their own family members, after they were told this was a famous con artist they were talking to. And most skilled con artists basically do not expect someone to be hard to manipulate, if not precisely in that extreme a circumstance.
So if Loki, god of lies and mischief, stabs an Umber Hulk and convinces it that actually it's someone over in that direction who stabbed them, so they go raging off in that direction busting through walls etc, while I don't particularly know how Loki achieved that, he was able to do that because he is very good at lying. Better than people in our world at lying, who are themselves using techniques they often struggle to describe and which most people who think they understand them (or claim to) are then unable to reproduce even slightly.
Lying is incredibly potent and powerful as a weapon, and most people are incredibly bad at it. So like most things PCs can do that their players can't (like cast Fireball), some handwaving is required to explain how they actually get from A to B (go on, cast fireball right now at the table to prove you know how your character does it - or it doesn't go off). Something like misdirecting an opponent as to who attacked them even if that person isn't present is the barest edge of what lies can accomplish and not impossible at all.
That said, if you asked whoever wrote that ability, they'd probably immediately clarify that they meant 'someone else adjacent' or 'someone else in the room'. They might even say they meant 'another enemy' of the creature targeted, and not say, an ally (getting accidentally stabbed by your allies in a confused melee is unfortunately quite common). It's very unlikely to be the intended use of this Rogue Talent, even though the text clearly makes it possible as written.
TLDR:
RAW - yes, that's how it works.
In-universe - lies are incredibly powerful. The GM may provide a penalty to your roll for an unbelievable lie, but provided you succeed, there is no reason this wouldn't work and I bet a skilled liar in our world could reproduce this while having less magical aids to his lying and less lies that work ('magical mind control' is less plausible in our setting).
RAI - the author likely did not think about that as an option and given pathfinder's relative level of power given to skills and rogues and fighters (as opposed to clerics, wizards and druids) would probably be shocked at the idea of rogues doing anything other than swinging daggers and picking locks.
* using human athletic world record holders as a baseline for maximum 'level' achievable in this setting.