Audit of the tone of communication: how to check compliance with the guide
What exactly to check if the guide is already written
The problem of most teams is not the lack of a guide, but the fact that verification reduces to a general feeling of "sounds normal / doesn't sound right." This is not a criterion, but guessing. A working audit is built on specific parameters, each of which can be checked binarily — whether it complies or not.
- Addressing the reader: "you" or "you" — and whether it changes within one text or between adjacent letters in one chain
- Degree of assertiveness of statements: the brand promises ("guarantee") or assumes ("may help", "in most cases") — mixing these two modes within one channel looks like a split personality of the brand
- Lexical layer: presence of words from the guide's "black list" (bureaucratic terms, competitor jargon, forbidden euphemisms) and presence of brand-specific formulations, if they are fixed
- Humor and lightness: is a joke allowed in this type of message (in an error payment notification — almost never, in a welcome letter — almost always)
- Length and rhythm of phrases: short chopped sentences versus expanded constructions — the deviation from the standard here is noticeable to the ear faster than it appears
These five points constitute the operational core of the audit. Everything else — syntax, punctuation, spelling — is the editor's job, not tone checking.
Frequency and sampling: how many texts actually need to be checked
Daily checking of each text — the path to team overload and to the audit being simply abandoned within a month. Another logic applies: frequency depends on the speed of content growth, not on the desire to 'do everything perfectly'.
| The situation in the team | Recommended frequency |
|---|---|
| A stable team, content volume is not growing | Once per quarter (90 days) |
| Scaling content, hiring new authors | Weekly |
| Startup with high employee turnover | Monthly |
The sample should not be large — sufficient is enough. The rule is simple: minimum two texts per channel and minimum two texts per each author or contractor, for a total of 5–10 fresh examples over the period. It is meaningless to check the entire array of texts: tonal drift manifests systematically, not in individual paragraphs, and it becomes apparent already from a small sample.
Case: how Southern Cross built an audience for 30+ channels
Southern Cross Health Insurance — an insurance company that faced a classic big brand problem: dozens of communication channels, from email to automated notifications, and each has its own author or contractor. Instead of a single text guideline for all situations, the company collected guidelines separately for each of 30+ channels, with each one supplemented with 'before and after' examples — not an abstract description of tone, but a live comparison of poor and good versions of the same message.
During the process, three zones were identified where the tone breaks most frequently:
- Transitions between channels — when a user receives an email in one tone, and then gets transferred to a chatbot or a page with completely different intonation
- Negative scenarios — refusals, errors, service restrictions: it is precisely here that authors most often fall into official language or, conversely, into inappropriate cheerfulness
- Automatic texts — notifications, bot texts, transactional emails, which are written once and then live for years without revision
The conclusion from this case is simple: audit cannot be limited to 'live' content from copywriters. Automated and systematic texts require a separate line in the review plan because by default no one edits them.
The tool for verification: a checklist, table, or AI service
The choice of tool depends not on the budget, but on the volume of content and whether the team works with external contractors or AI generation.
| Tool | When it fits | Limitation |
|---|---|---|
| Checklist (5 criteria from the section above) | A small team, manual review | Subjectivity in evaluation among different reviewers |
| Comparison table 'benchmark / text' | Working with contractors, documentation required | Requires time to fill out |
| AI service (Elysia Tools, ToneGuide, Brand Tone Checker) | Large volume of content, many authors | Does not replace fact-checking and context verification |
AI services like Brand Tone Checker evaluate text along four axes — Formal/Casual, Confident/Cautious, Optimistic/Concerned, Playful/Serious — and provide a numerical deviation from the benchmark. This is convenient for stream checking, but dangerous as a standalone tool: according to the BBC/EBU study, approximately 20% of neural network responses contain inaccurate or outdated information. If texts are generated or edited by AI, the fact-checking stage must be separate from tone checking — these are different tasks, and mixing them into one step is an error.
How to document results and provide feedback
An audit without recording results is useless — nobody will remember in a month what was fixed and why. The working practice is a table with three columns: a quote from the text, what is wrong according to the guideline criteria, and a revised version.
Example from real text work: the author wrote in a letter of refusal to refund «We guarantee that we will resolve your issue within 24 hours» — while the brand guideline prescribes cautious formulations in negative scenarios. Correction: «We usually resolve such issues within 24 hours» — the assertiveness is removed, and the tone returns to the benchmark range. Pairs of «was / became» like this should be collected in a separate document — over time, it becomes a living supplement to the guideline and good material for onboarding new authors.
Feedback should not be given as a list of errors, but rather with a specific criterion from the checklist that has been violated. "Too formal" is a useless comment. "Addressing through "you" instead of the accepted channel "thou"" is a comment that can be worked with.
How to catch voice drift before clients
Voice drift is not a one-time mistake, but a gradual shift that the team does not notice from within: each individual message seems normal, but after six months the brand tone differs significantly from the original. It needs to be monitored systematically, not relying on the editor's intuition.
The working model is a four-layer verification protocol, where each layer solves its own task and does not overlap with the others:
| Verification layer | Time | What they do |
|---|---|---|
| Weekly micro-check | 15 minutes | One item per channel — a quick snapshot of the state |
| Monthly audit of templates | 45 minutes | Checking automatic and recurring texts |
| Quarterly channel synchronization | 90 minutes | Tone verification between different channels and transition points |
| Annual audit by an external reviewer | Full day | A gaze without internal blindness, a complete guideline revision |
The main indicator that drift has already begun is when four markers from the criteria checklist start repeating in deviation simultaneously: the address changes, categoricality increases, foreign lexicon, humor appears, or disappears, or becomes inappropriate. If at least two of the four markers show deviation in several consecutive texts — this is not a coincidence, but a trend, and waiting for the quarterly audit is already too late. At this point, it is necessary to launch an unplanned weekly micro-check across all channels at once, without waiting for the scheduled date.
What exactly to check if the guide is already written
The problem of most teams is not the lack of a guide, but the fact that verification reduces to a general feeling of "sounds normal / doesn't sound right." This is not a criterion, but guessing. A working audit is built on specific parameters, each of which can be checked binarily — whether it complies or not.
- Addressing the reader: "you" or "you" — and whether it changes within one text or between adjacent letters in one chain
- Degree of assertiveness of statements: the brand promises ("guarantee") or assumes ("may help", "in most cases") — mixing these two modes within one channel looks like a split personality of the brand
- Lexical layer: presence of words from the guide's "black list" (bureaucratic terms, competitor jargon, forbidden euphemisms) and presence of brand-specific formulations, if they are fixed
- Humor and lightness: is a joke allowed in this type of message (in an error payment notification — almost never, in a welcome letter — almost always)
- Length and rhythm of phrases: short chopped sentences versus expanded constructions — the deviation from the standard here is noticeable to the ear faster than it appears
These five points constitute the operational core of the audit. Everything else — syntax, punctuation, spelling — is the editor's job, not tone checking.
Frequency and sampling: how many texts actually need to be checked
Daily checking of each text — the path to team overload and to the audit being simply abandoned within a month. Another logic applies: frequency depends on the speed of content growth, not on the desire to 'do everything perfectly'.
| The situation in the team | Recommended frequency |
|---|---|
| A stable team, content volume is not growing | Once per quarter (90 days) |
| Scaling content, hiring new authors | Weekly |
| Startup with high employee turnover | Monthly |
The sample should not be large — sufficient is enough. The rule is simple: minimum two texts per channel and minimum two texts per each author or contractor, for a total of 5–10 fresh examples over the period. It is meaningless to check the entire array of texts: tonal drift manifests systematically, not in individual paragraphs, and it becomes apparent already from a small sample.
Case: how Southern Cross built an audience for 30+ channels
Southern Cross Health Insurance — an insurance company that faced a classic big brand problem: dozens of communication channels, from email to automated notifications, and each has its own author or contractor. Instead of a single text guideline for all situations, the company collected guidelines separately for each of 30+ channels, with each one supplemented with 'before and after' examples — not an abstract description of tone, but a live comparison of poor and good versions of the same message.
During the process, three zones were identified where the tone breaks most frequently:
- Transitions between channels — when a user receives an email in one tone, and then gets transferred to a chatbot or a page with completely different intonation
- Negative scenarios — refusals, errors, service restrictions: it is precisely here that authors most often fall into official language or, conversely, into inappropriate cheerfulness
- Automatic texts — notifications, bot texts, transactional emails, which are written once and then live for years without revision
The conclusion from this case is simple: audit cannot be limited to 'live' content from copywriters. Automated and systematic texts require a separate line in the review plan because by default no one edits them.
The tool for verification: a checklist, table, or AI service
The choice of tool depends not on the budget, but on the volume of content and whether the team works with external contractors or AI generation.
| Tool | When it fits | Limitation |
|---|---|---|
| Checklist (5 criteria from the section above) | A small team, manual review | Subjectivity in evaluation among different reviewers |
| Comparison table 'benchmark / text' | Working with contractors, documentation required | Requires time to fill out |
| AI service (Elysia Tools, ToneGuide, Brand Tone Checker) | Large volume of content, many authors | Does not replace fact-checking and context verification |
AI services like Brand Tone Checker evaluate text along four axes — Formal/Casual, Confident/Cautious, Optimistic/Concerned, Playful/Serious — and provide a numerical deviation from the benchmark. This is convenient for stream checking, but dangerous as a standalone tool: according to the BBC/EBU study, approximately 20% of neural network responses contain inaccurate or outdated information. If texts are generated or edited by AI, the fact-checking stage must be separate from tone checking — these are different tasks, and mixing them into one step is an error.
How to document results and provide feedback
An audit without recording results is useless — nobody will remember in a month what was fixed and why. The working practice is a table with three columns: a quote from the text, what is wrong according to the guideline criteria, and a revised version.
Example from real text work: the author wrote in a letter of refusal to refund «We guarantee that we will resolve your issue within 24 hours» — while the brand guideline prescribes cautious formulations in negative scenarios. Correction: «We usually resolve such issues within 24 hours» — the assertiveness is removed, and the tone returns to the benchmark range. Pairs of «was / became» like this should be collected in a separate document — over time, it becomes a living supplement to the guideline and good material for onboarding new authors.
Feedback should not be given as a list of errors, but rather with a specific criterion from the checklist that has been violated. "Too formal" is a useless comment. "Addressing through "you" instead of the accepted channel "thou"" is a comment that can be worked with.
How to catch voice drift before clients
Voice drift is not a one-time mistake, but a gradual shift that the team does not notice from within: each individual message seems normal, but after six months the brand tone differs significantly from the original. It needs to be monitored systematically, not relying on the editor's intuition.
The working model is a four-layer verification protocol, where each layer solves its own task and does not overlap with the others:
| Verification layer | Time | What they do |
|---|---|---|
| Weekly micro-check | 15 minutes | One item per channel — a quick snapshot of the state |
| Monthly audit of templates | 45 minutes | Checking automatic and recurring texts |
| Quarterly channel synchronization | 90 minutes | Tone verification between different channels and transition points |
| Annual audit by an external reviewer | Full day | A gaze without internal blindness, a complete guideline revision |
The main indicator that drift has already begun is when four markers from the criteria checklist start repeating in deviation simultaneously: the address changes, categoricality increases, foreign lexicon, humor appears, or disappears, or becomes inappropriate. If at least two of the four markers show deviation in several consecutive texts — this is not a coincidence, but a trend, and waiting for the quarterly audit is already too late. At this point, it is necessary to launch an unplanned weekly micro-check across all channels at once, without waiting for the scheduled date.
Recent articles
MegaFon, BMW, VTB and more 240+ companies. See how it looks in the showcase reel:
Our branding agency is approached to order:
Branding agency services
