AI Brand Audit: A Template Branding Studios Can Sell

A studio can run a brand audit with AI by letting the assistant log touchpoints, extract competitor claims and check copy against the client's stated promise, while a person scores, ranks and recommends. The client deliverable should hold a one-page verdict, a scored touchpoint inventory, a competitor map, ranked findings and a 90-day action plan.

Many automated audit tools measure one slice, such as how a website reads or how assistants describe a brand online. A studio's audit also covers the invoice template and the sign over the door, and it ends in a plan. The audit below is a ten-working-day job for a music school with three sites and two logos. Every figure in it is illustrative, so replace the hours and fees with your own timesheets.

Follow me on Instagram@sagnikteaches

The brief: a music school that looks like two schools

The school has about 40 teachers and teaches at three sites. It was formed four years ago when two smaller schools merged, and it still shows both parents: a blue wordmark on some things, an orange symbol on others. The director's complaint fits in one line. Parents ask whether this is "the blue school or the orange school". She wants to know what to fix before she pays for a rebrand, and she wants the answer in two weeks.

Connect on LinkedInSagnik Bhattacharya

Before anything is scored, the studio fixes the yardstick, because an audit without one produces opinions about taste. It asks the director for a single sentence describing the school and gets this: "Every pupil performs, every term, whatever their level." Each touchpoint is then judged on whether it says that, under the agreed name, in a look and voice that match the others. If a client cannot write the sentence, run a short positioning exercise first, as described in the positioning and messaging workflow for agencies, and bill it separately.

Subscribe on YouTube@sagnikteaches

The split of labour is the part clients ask about, so state it early. The assistant logs, extracts, compares and drafts. People decide what matters to this client, how bad each finding is and what to recommend. Stick to that line and you can honestly call the audit yours. For how this fits the rest of a project, see where AI sits in a brand project.

Days 1 and 2: the touchpoint inventory, dull items included

The inventory is a spreadsheet with one row for every place the brand meets someone: a parent, a supplier, a job applicant. On day one the studio asks the director for three things: access to every channel, a folder of every document template and a photo walk of the three buildings. Insist on the dull items. The invoice template, the voicemail greeting and the teachers' lanyards are where a brand usually comes apart, because nobody owns them.

The school's list came to 41 touchpoints in seven groups.

What the assistant does here is read. Give it screenshots, photos and pasted text and it returns one structured row per item, which turns a day of typing into an hour of checking. Send images in batches of eight to ten, which keeps a wrong row easy to spot, and check your assistant's per-message attachment limit before the first batch, because plans differ. Use a business plan, or switch training off on a personal one, before you start, because the invoice template and the term letter are client documents.

You are logging brand touchpoints for an audit. For each image or text I send, return one CSV row with these columns:
id, touchpoint, group, logo (A = blue wordmark, B = orange symbol, C = other, none), name as written, first sentence, tone (max 3 words), visible date

Rules:
- Copy names and sentences exactly. Never paraphrase.
- If a field cannot be seen in the image, write "not visible". Do not guess.
- If an image is unreadable, return a row with the id and "unreadable" so I can reshoot it.
- Say which part of the logo you looked at: the wordmark text, the symbol, or both.

id,touchpoint,group,logo,name as written,first sentence,tone,visible date
W1,Home page,Web,B,"[School] Music","Lessons for every age and level.",warm; generic,not visible
D3,Invoice template,Documents,A,"[Old school A] Music Centre","Invoice for lessons",formal; cold,12 Mar
S2,Exterior sign (site 2),Premises,C,"[Old school B]",none,none,not visible

That sample output is illustrative, with placeholders where a real client name would sit. Three corrections came out of the school's first batch, and each is now a line in the prompt. The assistant treated two orange marks as the same logo because it compared colour rather than the symbol, so it must now say which part of the logo it examined. It filled the date column with plausible guesses, so anything a photograph cannot show must read "not visible". And it quietly skipped a faded sign photographed at an angle, which is why the studio counts rows against the shot list before moving on: 41 shots in, 41 rows out.

Days 3 and 4: a competitor scan that quotes instead of summarising

The scan covers five competitors: four direct rivals and one online platform that parents might choose instead. For each, save the home page, the page that sells the main service and the About page as PDFs, so you hold the wording as it stood on the capture date. Load the PDFs into a project or notebook. Gemini Notebook, the product formerly called NotebookLM, answers from the documents you load and attaches citations that link back to the passage, which suits this job. A ChatGPT or Claude project works too if the prompt demands exact quotes.

From these five competitors' saved pages only, list every claim each one makes about itself.
For each claim give: competitor, the exact quote in quotation marks, the file it came from.
Then group the claims into themes (teachers, ages and levels, exams, performance, progress, price, convenience).
Do not infer anything that is not written on the pages. If a theme has no claims, say "none found".

Every claim then goes into a frequency table that counts how many rivals make it and how many of the client's own touchpoints state it. On the school job the table looked like this.

Read the last two columns together. A claim that all five rivals make cannot set the school apart, however true it is. The one claim no rival makes appears on only 3 of the school's 41 touchpoints: a buried sub-page, an Instagram highlight and a line in the prospectus. That finding became the spine of the report. The school's strongest promise is its least visible one.

Check before you trust any of it. Open each quoted line in the saved PDF and delete every quote you cannot find word for word. In this job the assistant credited one rival with an "exam-focused" positioning, and the line came from a two-year-old news post that had been saved by mistake, not from a page that sells lessons today. Record the file name and capture date beside every quote, because that record is what makes the table defensible in front of the director.

Days 5 and 6: scoring with a rubric an assistant cannot flatter

Scoring is where audits slide into opinion, so the rubric carries the weight. Each touchpoint gets 0, 1 or 2 on five criteria, a maximum of ten. The anchors below stop the assistant, and the studio's own mood on a Friday afternoon, from drifting.

Voice needs the most care, because it is easy to wave at and hard to score. The agreed voice here is "warm, plain, specific", lifted from the director's sentence. How agencies keep brand voice consistent in AI-written content shows how to turn a voice description into rules a model can follow, which is also what makes the 0-1-2 anchors scoreable.

Run the scoring in a fresh chat that holds only the rubric, the yardstick sentence and the evidence for each touchpoint. Keep the director's opinion of her own brand out of it. If you doubt that matters, test it: score the same invoice template once with "this is a strong brand" in the prompt and once without, and compare the totals. Ask for a quotation that justifies every point awarded, and discard any score whose quotation is not in the evidence you supplied.

Score each touchpoint below against the rubric (Name, Look, Voice, Promise, Proof; 0, 1 or 2 each).
Yardstick sentence: "Every pupil performs, every term, whatever their level."
Agreed full name: [School] Music School. Agreed voice: warm, plain, specific.
For every point you award or withhold, quote the words or describe the visible feature that justifies it.
If the evidence for a criterion is missing, write "no evidence supplied" and do not score it.
Return a table: id, name, look, voice, promise, proof, total, evidence.

That last rule exists because of a mistake from the school job. The first run scored the voicemail greeting 0 out of 10: no transcript had been supplied and the assistant read missing input as failure. Once the studio typed the greeting out, it scored 6. A second slip went the other way. The enrolment form scored 9 because its heading carried the right name, while the footer still named one of the old schools. The studio's sampling rule caught it: a person reads every touchpoint scoring under 5 and one in five of the rest, and overrides any score whose evidence does not hold. Write that rule into the method page of the report so the client can see how far to trust the totals.

The website scores best because it was rebuilt after the merger; the signs, printed items and phone greetings were not. Two further counts came straight from the inventory, not from the assistant. Four versions of the logo were live: the blue wordmark on 12 touchpoints, the orange symbol on 17, two older marks on 5, and none on 7. Three colour palettes were in use. Add the columns in the spreadsheet and never ask the assistant to total them, because models can miscount a long table and a wrong sum on page 4 undermines the other eleven pages.

Day 7: what the assistants say when a parent asks

Some parents will skip search and ask an assistant which music school to try for an eight-year-old. The studio writes eight questions of that kind (best school for beginners, lesson prices, how performances work, who to contact), asks each one in three assistants twice, and logs 48 answers. For each it records whether the school is named, whether every fact is right and which sources are cited. In this job the school was named in 11 of the 48 answers. Five of those 11 carried a wrong fact, such as an old school name or a site that had closed. None mentioned the termly showcases.

Treat the result as a dated snapshot. Answers vary from run to run and between accounts, which is why each question is asked twice rather than once. The remedy is not clever optimisation. It is making the school's own pages and its three listings say the same thing, so every source an assistant might draw on agrees. For a one-off audit, 48 hand-run answers are enough. Paid trackers exist for monthly monitoring (one lists a one-brand plan at $49.95 a month as of November 2026), and monitoring brand mentions in ChatGPT for a client covers that routine if the client wants it to continue after the audit.

Days 8 and 9: from 31 findings to six ranked recommendations

By day eight the working file holds 31 findings: 17 from the scoring, 6 from the competitor scan, 5 from the assistants' answers and 3 from the photo walk. The assistant's job is to cluster duplicates and draft wording. The studio's job is deciding which findings matter to this client. The difference shows in how one finding changes between the assistant's draft and the version that goes in the report.

The studio version carries a count, a plain picture of what the parent sees, an owner for the decision and a deadline. Rank recommendations by reach (how many touchpoints, or how many parents, the fix touches), then severity (how far the problem contradicts the promise), then effort. Anything that depends on a decision, such as replacing signs, waits for that decision.

Recommendation 1 is where the studio's identity work begins, and what AI can and cannot do for logos, palettes and guidelines sets out which parts of that job still need a designer. The audit's role is to make the decision easy to take. Picking blue means touching the 17 orange touchpoints, picking orange means touching the 12 blue ones, and 12 more need work either way.

The twelve pages the school receives

This is the part a studio can package and sell, so it is worth being strict about what goes where. The assistant drafts some pages; every page has a named human owner.

Two rules do most of the work. Page 1 must stand alone, because the director will forward it to her board without the other eleven. And every claim on pages 4 to 11 carries an evidence ID from page 12, so any score can be challenged and defended. Leave out the raw assistant transcripts, any "brand health" number you cannot explain in a sentence, and design opinions with no evidence behind them.

The same method for a kitchen fitter and a care agency

The groups change with the business, but the method does not. A kitchen fitter's highest-reach touchpoint is probably the van, followed by the quote document and the snagging form that customers see on the last day. Proof matters most there: dated photographs of finished kitchens score 2, a line saying "quality workmanship" scores 1. The photo walk is a depot visit and an hour in the car park, not a trip round three music rooms.

A care agency's inventory leans on documents: the care-plan template, visit logs, the welcome letter and recruitment adverts, plus carers' uniforms and badges. Voice carries more weight because messages reach families at stressful moments. Proof must be checkable, so any inspection rating quoted on a touchpoint should be quoted in full with its date, or left off. The studio adds a column for "who reads this and in what state of mind", which changes how severe a cold or vague sentence really is.

Packages, price logic and the scope wording

Selling the audit means fixing what a client gets before kick-off. Three packages cover most requests. The hours below are planning figures for an AI-assisted studio; replace them with your own timesheets after the first two jobs.

The Extended package leans on interview synthesis, which turning interview notes into a findings deck covers in detail. Here is a quick sum for the Standard package. The school job took 28 hours; the same work unaided would have taken about 51 (inventory 10, competitors 9, scoring 12, assistant check 6, report 14). At an internal cost of $90 an hour that is $2,520 against $4,590. A studio charging a fixed $4,800 keeps about $2,280 with AI and about $210 without. AI changes the studio's margin, not the price the client sees, and the price should rest on the decision the audit prevents, not on the hours.

Put the scope in writing in words like these, and adapt them with your own lawyer or adviser.

The audit covers the touchpoints the client can show us within five working days of kick-off, up to 60. Anything shown later is recorded as "not reviewed". Scores measure consistency with the positioning sentence the client approved on day one. They are not a measure of commercial performance. We use AI assistants to log and compare material; a named member of our team has checked every score, quotation and recommendation in this report.

What the studio changed before its second audit

Three changes came out of the school job. The studio now asks for the document folder on day one, because the templates arrived on day six and pushed scoring back by a day. It shows the director the scorecard before the findings are written, because her reaction to the 47% for print and premises decided which findings the report led with. And it starts the assistant check on day three, since two sessions on different days are the cheapest way to see how much an answer moves. None of these changes touched the method; they only moved when the client sees things.

Checks before the report leaves the studio

• Every quote traced. Each competitor quote has a file name and capture date and has been found word for word in the saved page.
• Every score evidenced. Each point maps to an evidence ID on page 12; anything marked "no evidence supplied" is either fixed or shown as not scored.
• Sums done in the spreadsheet. Totals, counts and percentages come from formulas, not from the assistant.
• Settings checked. Client files only went into business plans, which do not train on your content by default, or into personal accounts with the training setting switched off. ChatGPT Business and Claude Team both need at least two seats, so a one-person studio may be on a personal plan.
• Freshness. Competitor pages re-opened within seven days of the report date, in case a headline changed.
• Yardstick signed off. The director has approved the positioning sentence in writing, so nobody can argue afterwards that the audit measured the wrong thing.
• Disclosure included. Page 2 says how AI was used and who checked the output.

Run through that list once as the person who built the audit and once as the director reading page 1 on her phone. If page 1 does not make sense without the rest, rewrite it before anything else leaves the building.

Questions studios ask before selling an AI-assisted audit

Which AI should a studio use for a brand audit?

Any of the main assistants (ChatGPT, Claude or Gemini) can read screenshots and pasted copy, so choose on data terms and file limits rather than brand. Business plans such as ChatGPT Business and Claude Team do not train on your content by default, though both need two seats. On a personal plan, switch the training setting off before you load any client material.

How many touchpoints make a full audit?

Thirty to sixty covers most small businesses. Stop adding items when new ones only repeat existing findings, and make sure every group appears at least once: web and listings, social, documents, print, premises, paid and recruitment, voice. Write the cap into the scope so a client cannot keep adding items after kick-off.

Should the client see the assistant's raw output?

No. Share the scored inventory, the quotes with their sources and your recommendations. Raw output holds unchecked claims and invites arguments about wording instead of findings. Do add a short statement of how AI was used and who checked the scores, because clients increasingly ask and being upfront costs nothing.

How is this different from an automated brand audit tool?

Automated tools tend to look at one slice, such as how a website reads or how assistants describe a brand online. This audit also covers print, premises, documents and voice, ties every score to evidence and ends in ranked recommendations. The tools are a useful input; the judgement and the plan are what a client pays a studio for.

Further reads

• AI Brand Naming: A Studio Workflow With Domain and Trademark Checks — The naming stage that often follows an audit, with domain and trademark screening.
• How to Use AI for Graphic Design: Brand Kits and Templates for Clients — Turn audit findings into brand kits and templates a client can use.
• How to Build a Brand Voice Guide That AI Can Follow — Write down the voice you scored against so models can follow it.
• AI Creative Brief Generator for Agencies: From Client Call to Brief — Convert the audit workshop into a creative brief the team can work from.
• Best AI Brand Visibility Software: Is AI Mentioning Your Business? — Compare trackers that check how assistants mention a brand each month.
• AI Consulting Report Writing: From Findings to a Client-Ready Draft — Structure long client reports from findings to a clean first draft.
• Best AI Tools for Creative Agencies: Briefs, Concepts and Client Sign-Off — See which tools help from brief to approved concept, and where they slow work.
• PPC Audit Template: Run Google Ads Audits for New Clients With AI — A five-section Google Ads audit template, the AI prompts that summarise the exports, and a filled sample with a one-page prospect summary.
• Design Studios Using AI: Briefs, Moodboards and Client Revisions — A studio-side workflow for moving from a messy brand brief to approved visual directions and controlled revision rounds.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google's Gemini Notebook help pages, the ChatGPT Business and Claude pricing pages, Google Workspace's AI privacy page and an AI-visibility tracker's pricing page, all read in November 2026.