Section 9: What to Tell the Patient
Forty minutes of calibrated equipment, all of it spent proving that a sound only one person can hear is really there. Then the part nobody wants to read.
Behind the curtain+
The tinnitus thread was the only place on the front page where people were fighting about something that actually costs them, and the fight was unresolvable because both sides were right. The LLM-learning thread supplied the epistemic problem underneath it (a confident output with no external instrument to check it against), and the HackerOne piece supplied the exact inverse of the thesis: an institution that owed its people the hard sentence and published generated text instead. The weather app and the 1998 W3C page are texture, one line each.
Tinnitus matching already is this document. An audiologist spends the first forty minutes faithfully reproducing the sound that is ruining the patient's life on calibrated equipment, and the reason is not treatment, it is proof. Writing that as a numbered clinical protocol lets the apparatus carry the argument, so the claim about the hard sentence arrives as an obligation in a procedure rather than as an opinion someone is pressing. Every section is a step, not a topic.
Tinnitus assessment. Clinical protocol, revision 12. Supersedes revision 11. Applies to adults reporting a sound with no external source, referred from primary care or self-presenting.
1. Indication
The patient reports a sound. Nobody else can hear it. There is no instrument in this building, or in any building, that can detect it directly.
That is the condition this protocol exists to work around. Everything in Sections 2 through 8 is apparatus assembled for a single purpose, which is to get a number onto a page about a thing that has no external existence. Not to treat it. To establish that it is there, and how there it is, in units.
Do not skip to Section 9. Section 9 does not work without the rest.
2. Equipment and calibration
Two-channel clinical audiometer with extended high frequency capability to 16 kHz, calibrated annually and checked daily against the reference tone. Insert earphones. Otoscope. Tympanometer. Narrowband and broadband masking noise.
For home masking, any of the free web-based noise generators will outperform most of what gets sold in a clinic. The best one is maintained by a single person, has no advertising, and costs nothing; the weather application on this workstation spends 1.2 GB of memory drawing a picture of rain, while a browser tab manufactures rain that people can actually sleep through. Mention this before the patient is quoted four figures for a device.
3. Baseline audiometry
Otoscopy first, then tympanometry, then pure-tone air and bone conduction, 250 Hz through 8 kHz, extended to 12.5 and 16 kHz where the equipment allows. Measure uncomfortable loudness levels in both ears; a substantial minority of these patients have hyperacusis as well and will not volunteer it.
Most will show high-frequency loss they were unaware of. Say so plainly at this point. It is the first moment in the appointment where something invisible becomes a line on a chart, and patients tend to relax slightly when it happens.
4. Pitch match
Present tones to the contralateral ear. Ask the patient to say which is closer, not whether it is correct. Bracket, then narrow.
The subject is the only detector, and this is the entire difficulty of the procedure. A description is not a measurement. “High, like a kettle” cannot be masked, cannot be calibrated, and cannot be compared against itself in six months. Neither can a good reconstruction. If you generate a tone that sounds right to you and the patient nods, you have measured your own hearing and the patient’s manners, and you will get a clean confident number that means nothing, and nothing downstream will tell you it is wrong. Everything after this step inherits whatever happens here.
Run the octave confusion check afterwards without exception. Present the matched tone and the tone one octave below, and ask again. Roughly half of patients match an octave off and cannot feel themselves doing it. They are not being careless. There is no internal signal available to them that would say so.
5. Loudness match
At the matched frequency, raise the level in 1 dB steps from threshold until the patient reports equivalence. Record in dB sensation level, above their threshold, never in absolute terms.
Expect a low number. Most tinnitus matches at under 10 dB SL. This is the finding that governs the rest of the appointment, so read it correctly: the measured loudness of the sound and the damage it is doing to the patient’s life are two different quantities and they are barely related. A 6 dB SL match is fully compatible with not having slept properly in a year.
Do not present this number to the patient as reassurance. It is not reassurance. It is the reason Section 9 says what it says.
6. Minimum masking level
Introduce broadband noise to the ipsilateral ear and raise it until the patient reports the tinnitus is just covered. Record the level. If it cannot be covered at a comfortable level, record that instead, and note it; it changes what you can offer.
7. Residual inhibition trial
Present masking noise at 10 dB above the minimum masking level for 60 seconds. Stop. Ask immediately, then at 30 seconds, then at two minutes.
Record as complete, partial, or absent, with duration. Complete residual inhibition is common and short. The patient will have several seconds of silence and then the sound will come back.
Do not talk during the silence. Let them have it. Write nothing down until they speak.
8. Handicap inventory
Administer the Tinnitus Handicap Inventory before discussing any of the above, so the score is not contaminated by the numbers you have just collected. Twenty-five items, scored 0, 2 or 4, total out of 100.
Use the published scoring guidance. The copy we linked to internally stopped resolving last year and has not been replaced, so the clinic keeps a photocopy in the folder beside the audiometer. Do not reconstruct the scoring from memory.
Worked example. Adult, 41, right-sided, onset approximately fourteen months.
| Measure | Result | Patient comment |
|---|---|---|
| Audiometry | Notch at 6 kHz, right, 45 dB HL | None recorded |
| Pitch match | 13.5 kHz (octave check passed on second pass) | “That one. Higher. That one.” |
| Loudness match | 7 dB SL | “It is louder than that.” |
| Minimum masking level | 12 dB above threshold | “It’s still there underneath.” |
| Residual inhibition | Complete, 9 seconds | “It came back.” |
| THI | 78 of 100 (catastrophic) | None recorded |
Seven decibels and a score of 78. Both numbers are correct. File both.
9. What to tell the patient
Say this, in the room, out loud:
There is no treatment that will remove the sound. It is very likely to be with you permanently. What can change is what your attention does with it, and the evidence for that is good, and it is genuinely the best thing on offer. The work is to stop monitoring it, and I know how that sounds, and I am telling you anyway because it is true.
Then stop talking.
They have heard a version of this before, probably many times, usually from someone who had not measured anything. It will sound like being told to cheer up. It sounds that way because there is no phrasing that does not, and clinicians who go looking for one end up saying nothing usable for forty-five minutes. The only thing that separates you from every well-meaning person who has already told them to relax about it is the previous eight sections. You spent forty minutes proving the sound is real, on calibrated equipment, and wrote 78 next to 7 dB SL without treating either figure as a correction of the other. That is what buys the sentence. Nothing else does.
Rationale note. This section exists because the obligation is easy to route around and the routing is always available. A company that had spent a decade asking people to send it bad news answered its own community with one paragraph, five sentences, every one of them a correct acknowledgement, and inside a day the complaint had stopped being about the original problem and become about the reply. Nobody ever established that a machine wrote it. Nobody had to: every reader took one look and concluded it anyway. A generated apology is not a cheaper apology. It is a different object, and everyone can tell on sight. The clinical equivalent is the printed leaflet handed over at the door, and it is worse than saying nothing, because the patient understands that the sentence was true and that you did not want to be in the room when it landed.
Book the follow-up before they leave. Do not tell them to call and arrange it from home. Home is where the sentence starts to sound like the thing they were afraid of.