How I picked
the organic content topics
for Nuud.

The wider strategy was already clear, which was to reduce Nuud's reliance on paid acquisition by building organic and AI-discoverable presence. This case study is really the layer underneath that, particularly because before any actual content could be commissioned for Instagram or TikTok, I had to answer a question the strategy assumed but didn't really solve, which was what the content was actually going to be about. I would say picking the wrong topics would have burned the limited content production capacity we had on posts nobody really searched for, and indeed it would have done so in a way that wouldn't really have shown up as a failure for months.

The brief I set myself was reasonably specific. The new social strategy needed awareness and consideration content that could pull people into the funnel at a lower CPM than paid search, and that read as useful rather than as advertising. Awareness in particular was the gap, particularly because Nuud at that point had effectively no top-of-funnel content at all. So the topics had to relate to natural deodorant in spirit, but couldn't really feel like deodorant adverts, and ultimately they needed to live in the adjacent territory of sweat, skin and armpit care, where I would say there is actually genuine curiosity.

While I already had some hunches about which topics could work, I wanted to use the same validation method I'd used during my York University co-op at Phoenix Agency, which is verified data from multiple sources cross-checked against each other, before actually committing any of the content budget. The challenge however was that Instagram and TikTok don't really publish search data the way SEO and SEM platforms do (there is no "search volume" column for Instagram), and so the work was as much about constructing demand signal where the platforms didn't really give it freely, as it was about picking the right keywords.

The narrative below is really the synthesised version. The 101-page working log behind it is embedded throughout this case study at the moment each section actually references it, so that every screenshot of Google Trends, Answerthepublic, Semrush, the Audit Atlas and the TikTok Audit Explorer is shown on the actual page where I was making the decision it informed. Feel free to open any embed and scroll through the surrounding pages of the log if you want the raw context.

I wrote these down at the start of the project on purpose, because every method I added afterwards had to actually map back to one of them. If a tool or step didn't help answer one of these questions, I didn't really see the point of putting it in the workflow.

I started with four candidate clusters, weighted toward awareness particularly because awareness was the gap. Three of them sit fairly close to the product. The fourth one (natural skincare) sits one ring out, which is broader and more crowded, but it has a much larger audience and a real adjacency to Nuud's positioning. Including it meant accepting that we would have to filter fairly aggressively for relevance later on, while excluding it would have meant capping the addressable audience too early. I would say keeping it in was the better call.

To keep the dataset legible as it grew, I built a single Google Sheet that was structured by cluster from the start, so that every term added later, from whatever source, would inherit a known category. Three columns at the beginning (prompt name, prompt cluster category, prompt source), and then two more added later for per-platform traffic and per-platform hashtag volume. I tried to keep the spreadsheet simple enough that a junior analyst could pick it up cold.

I ran Google Trends first particularly because it's free and fast, and it tells me whether a cluster actually has any meaningful interest before I commit time to recursive expansion. The pattern I followed per cluster was fairly simple: drop the seed term in, look at "Interest over time" for stable or rising lines (and indeed flat-at-zero clusters get dropped on the spot), and then read the Top queries and Rising queries lists for sub-terms that might be worth promoting into the master sheet.

The first cluster I ran was "natural deodorant ingredients." Interest over time was reasonable, not spike-heavy, but with a stable baseline and a slight rising direction over the past year. The Top queries table surfaced "best natural deodorant" and "aluminum free deodorant" as the two strongest related terms, and the Rising queries table promoted "aluminum free deodorant" again (+160%) and "best natural deodorant" (+40%). Both of those went into the sheet, with the awareness/consideration framing noted next to each.

The second cluster ("how does sweat work") looked more interesting than I had expected. Interest over time was lower than the deodorant-ingredients line at baseline, but the Top queries table surfaced "what is sweat" with high search interest and a +2% climb, and "how does sweat cool the body" rising at +4%. I would say both of those landed in the awareness bucket on the basis that an explainer video about either is content the audience is probably not going to read as advertising. They both went into the sheet.

The third cluster ("armpit care / armpit health") is really where the methodology had to split. I had already split the cluster pre-Trends because I knew the health side was going to intersect EU advertising rules, and Trends confirmed it pretty quickly. The Top queries list on "armpit health" was dominated by urgent care, armpit lump, and lymph nodes, and every one of those is either a medical query I couldn't really address or a non-buyer query I have no business teaching about. The Rising queries even included "Dove advanced care deodorant" (BREAKOUT) and "hidradenitis suppurativa", one being a competitor, the other being a medical condition. So I promoted nothing from the Trends pass on this particular cluster, and decided to retest it later via Answerthepublic instead.

The spider-web idea is reasonably simple. A seed term gets fed into Answerthepublic, and each related query that has identifiable traffic then becomes the next seed. Each of those produces its own children, and so on, until the branches stop returning any traffic. The output is really a dataset that ends up much wider than any single keyword tool would offer, and it gets weighted fairly naturally toward terms that actually have audience demand rather than just theoretical demand.

To demonstrate the method I ran the first term ("natural deodorant ingredients") fully manually. Answerthepublic returned a search volume of 1.6K on the seed term itself, and that value went straight into the Traffic ATP Google column in the master sheet, against the row that had been a placeholder seed about five minutes earlier, and indeed that was really the first actual data point of the dataset.

The next view down (Answerthepublic's "AI Prompts" panel) surfaced a long list of natural-language questions people ask AI tools about deodorant ingredients, things like "What are the most effective natural deodorant ingredients for sensitive skin?", "What are common ingredients in aluminum-free deodorants?", or "Which natural deodorant ingredients are best for odor control?". These really represent ideas, not validated demand, particularly because there is no traffic data or platform breakdown attached, so I specifically didn't promote them yet. They got noted and parked for later cross-checking against actual search data.

The Keywords panel below the AI Prompts list did have volume numbers attached. "Natural ingredients deodorant for men" came in at volume 10, light, but real demand, so it went into the sheet. The "people also ask" mind map that came next was probably the most useful single screen in the Answerthepublic flow, really a visual spider web of the question-shaped queries surrounding the seed term, with sub-questions branching out from each one. Most of the medical-adjacent branches I excluded immediately (the EU rule again), and "What ingredients are in natural deodorant" was the only one I really promoted, particularly on awareness-stage logic.

One term took me about twenty minutes to expand manually, and there were going to be dozens of them to run, plus their children. At that pace I would say the spider-web phase would have consumed about two weeks of clock time I didn't really have, particularly on work that is almost entirely data-shaping rather than actual judgement. So I switched the pattern. I downloaded the Answerthepublic exports for each new seed term, opened a dedicated Claude project, gave it the rules sheet, and then handed it the exports in batches.

Once the batching pattern was actually working, the dataset grew fairly quickly. The screenshot below shows the master sheet after the first round of batched expansion, with clusters intact, source attribution preserved, and the per-platform traffic columns starting to populate. Below the main list is a secondary block of Instagram-specific hashtags that Answerthepublic surfaced as having post volume or post count attached, which I kept on a separate sheet specifically so they wouldn't be confused with query-level entries.

With the spider web returning more terms than I could reasonably commission content for, I needed a stronger filter than just "has some traffic." Inside the Semrush free-tier limit (there was no company budget for the paid version), I sorted the master sheet by Google ATP traffic and took the top 20 terms. Each one I entered manually into Semrush, with two columns added per row for the actual Semrush volume and the KD (Keyword Difficulty). The work was split across two days specifically to respect the daily search cap.

The worked example I ran manually was "natural deodorant ingredients." Semrush returned a global volume of 1.0K (US 590, with UK/AE/AU/CA/CH each contributing 20–30), a KD of 26% (categorised as "Easy"), and informational intent. That is really an easy-to-reach Google surface with informational intent, which I would say means a content-led page can probably win it without a backlink war. A KD of 26% on a category-relevant term was probably the strongest single data point of the entire SEO pass.

Before running the Semrush pass I had to add two new columns to the master sheet (Traffic Semrush and KD score Semrush), particularly so that the data would have somewhere to actually land in the same row as the rest of the per-platform traffic. The sheet now had eleven columns and was getting fairly wide. To keep it usable I sorted by Traffic ATP Google descending, which put the highest-traffic terms at the top, and that was really the set I wanted to query Semrush against first inside the daily search cap.

The sheet below page 14 in the log shows the manual top-20 entry in progress. I worked the rows from the top down, opening each term in Semrush, screen-capturing the keyword overview, and copying volume + KD into the right cells. The terms that survived this filter, informational intent, KD < 35, decent volume, became the priority list for the manual social-platform audit that followed. The terms that flunked the filter stayed in the sheet, just unstarred. They might survive a later re-cut after the audit shows which queries actually generate engagement, and I didn't want to delete them and lose the lineage.

This is really the step that hand-built the dataset Answerthepublic refused to give me. For each of the 30 highest-traffic surviving queries, I opened the Instagram app and recorded the top 10 results, with columns for post name, creator, total engagement, views (where visible), page followers, engagement % by views, engagement % by followers, and query source/type.

The first armpit-care query I ran manually before I had Claude in the loop, and the result on its own kind of reframed how I read every later report. The results grid for "armpit care" on Instagram was full of videos with 7K, 110K, 209K view counts, and yet the tools had been reporting this cluster as "0 volume." The screenshot from inside the Instagram app at 20:23 that evening, showing the search-results grid for "armpit care" with view-count overlays in the bottom-left of each thumbnail, is really the moment the audit's methodology hardened for me, particularly because the tools were under-reporting niche social by what I would say is an order of magnitude.

Inside that grid, the post I pulled first was @glowjournalmamata's "We don't judge, bright pits only", which was captioned as a paid partnership. The engagement overlay on the post showed 353 likes, 48 comments, 3 shares, and 142 saves, for a total engagement of 544. The view count on the next post (same thumbnail style) was 93,900. The next step from there was the creator profile itself, which showed 267 posts, 50,700 followers, and a Melbourne-based digital creator in skincare/beauty.

Below that pair of screenshots in the log is really the first row of the manual-audit spreadsheet, populated for the very first time. Query name "Armpit care", creator "Glowjournalmamata", post name "We don't judge, bright pits only", total engagement 544, views 93,900, page followers 50,700, engagement % by views 0.57%, engagement % by page followers 1.07%, query source/type "Original cluster". Those nine fields really became the schema for every later capture across both platforms.

Doing this fully manually would have taken days I didn't really have. The workaround that ended up unlocking the rest of the project was to take a paired screenshot of the post and the creator profile, hand both to Claude inside a project, and ask Claude to read the metric overlays directly into the spreadsheet. The model isn't really doing OCR in the formal sense (it's more vision-reading numbers off the screenshot), but for this particular dataset the error rate was fairly low, and the time saved was, I would say, an order of magnitude.

The first ten-post batch for "armpit care" makes the cost-per-row reasonably tangible. The top performer in this batch was @minseon.kimm's "How to get CLEAN ARMPITS" at 11,616 engagement, 209K views, 72,800 followers, which is a 5.56% engagement-by-views rate and 15.96% engagement-by-followers. The bottom of the batch in the same query was @waneetacantik's "My Basic Armpit Care" at 45 engagement, 4,043 views, and 351,000 followers, which is a 0.01% follower-engagement rate. I would say the variance inside a single query was always pretty wide, and indeed that's really why the variance-shape rule from the next chapter ended up in the workflow.

The full Instagram dataset closed out at around 159 audited posts across 18 queries inside 5 clusters. The columns visible in the screenshot below are query name, creator handle, post name (truncated), total engagement, views, page followers, engagement % by views, and engagement % by followers. The dataset is sortable by every dimension (query, creator, engagement, view count), which is really what made it useful as the input for the searchable HTML explorer that came next.

~159 rows is enough to actually be useful and too many to scan in a spreadsheet, so I asked Claude to convert it into a small searchable HTML explorer (which I called Audit Atlas) so that anyone on the team could open it, filter by query or cluster, sort by engagement, and get to the underlying posts and metrics in about two clicks. The tool wasn't really supposed to be pretty. It was supposed to outlive me on the brand and stay readable to whoever inherited the channel later on.

The header carries the headline counts (159 posts, 18 queries, 5 clusters) and two tabs: Popular Posts (default) and Popular Queries. The search bar accepts free-text matches on creator, post or topic, and there is a filter dropdown to constrain by Cluster or Query, and a Sort dropdown that defaults to Engagement descending. The top of the first view shows the highest-engagement single post in the entire dataset, which is @yogiwhispers' "Routine for Breast Health" at 101K engagement, 1.8M views, 366K reach, and a 27.51% reach efficiency. The query that surfaced it was "armpit health."

The Popular Queries tab is really where the strategic decisions actually got made. Each row is a query, aggregated across all posts captured for it, with columns for total engagement, post count, creator count, average engagement, median engagement, total views, and the top post within that query. The dataset's leader by total engagement was "natural skincare", which came in at 107K total engagement across 10 posts and 10 creators, averaging 11K per post, with a median of 1.3K, and total views of 4.6M. The top performing post in that query was @ashikaroy_._ at 88K engagement, which is a DIY piece on "Dark circles, puffy under-eyes".

The second-most-engaged query was "armpit health" at 104K total engagement across 10 posts and 9 creators. But the per-post variance here is wider, with the top post at 101K and the median at 37. I would say the distribution shape on that one mattered more than the topline number, particularly because a high-mean, low-median query is really one where a couple of viral posts carry the average while most of the other posts flop, whereas a high-mean, high-median query is more of a category where competent content reliably performs. "Armpit health" turned out to be the former, which means entering it really means betting on going viral, and I would say that's not really a strategy you can hand to an intern.

Within the "natural skincare" query, scrolling past the top result, there were two posts that gave me immediately reusable formats. @camy.hati's "How is your skin so healthy? / Everyday choices make the biggest difference" (4.7K engagement, 786K reach, 0.60%), particularly the hook there, translates directly into "How are your armpits so healthy?" with a Nuud USP swap at the consideration stage. And @victoria_benitez's "If it is 'harmful if consumed' it's not skin care. It's poison" (4.0K engagement, 496K reach, 0.81%), I would say is a fairly strong awareness-stage hook for the natural-ingredients angle that we already lead on. Both of those went straight into the hook bank.

Third by engagement in the Atlas: #naturalskincareproducts at 35K total engagement across 10 posts, 9 creators, 316 median engagement, 807K total views. Top performing post: @thrivewithcandicee at 19K engagement on "I haven't washed my face in over 2 years and it's healed my skin / My super minimal & nontoxic skincare", against 737K views and 597K reach, a 2.61% engagement-by-views rate.

This query mattered to me not really because of the topical relevance (we don't sell face wash), but more because the format is really a near-perfect twin of one of Nuud's strongest credible claims, which is "I haven't applied deodorant in X hours and I smell less". Both posts trade on a counter-intuitive personal statement that the audience kind of wants to know the explanation for, and indeed the hook engineering is identical, with only the noun changing. So that format twin went straight into the consideration-stage brief.

The seventh query, "what is sweat", had 12K total engagement, 5 posts, 5 creators, 2.0K median engagement, 543K total views. That's a small dataset and a niche topic, but it ranked seventh on engagement, which on a five-post sample means each post was punching. The top performer, @iamyenlikethemoney's "Vietnamese Salted Coffee / Salt coffee – cà phê muối" at 7.5K engagement, was tangential to the query, but the trend across the rest of the posts was clear: educational, body-systems content earns engagement on Instagram even when the query is text-book-sounding.

The tenth Audit Atlas query, #aluminumfreedeodorant, is the one that matters most for Nuud, because aluminium-free is one of our hard USPs and the hashtag itself filters straight to category-fluent intent. The query produced 6.6K total engagement across 10 posts, 9 creators, 58 median engagement, 233K total views. Top performer: @madisonbrown.pac's "You still putting aluminum underneath your armpits? / I stopped using antiperspirant" at 4.5K engagement, 62K views, 145K reach, a 7.30% engagement-by-views rate, one of the highest in the entire dataset.

That post is excellent content-wise and format-wise: the shock-hook framing ("why are you still putting this on?") plays into the consideration stage of our funnel, and the structure is one we can directly replicate in our own voice. The second flagged post, @melissa_gandarinho's "I may have found the best aluminum free deodorant! / You can use my ShopMy link" at 20 engagement against 492 views, is the other end of the same query, showing how the "best" framing flops when it reads as a sponsorship. The brief that came out of this query: shock-hook the USP, never lead with "best."

TikTok really behaves nothing like Instagram, and the audit had to be re-scoped from the very first day on it. The Creative Center is probably the cleanest entry point for it, particularly because it shows top ads by category, format and time window, with retention curves and CTR bands per video. The goals I had for this stage were format inspiration, identifying category-level trending terms (with the filter set to "Beauty & Personal Care" and "Last 120 days" to filter out short-term noise), and then running the same Answerthepublic + manual audit pattern from Instagram on the TikTok side too.

Searching "armpit care" on Creative Center actually returned skincare ads that were largely unrelated to deodorant, things like financial planning and plant care, and so none of it was particularly useful. "Armpit health" did somewhat better. The first relevant result on that latter query was a skincare ad titled "Why are antioxidants important for skin health?" with a Top 89% CTR, low budget, region "Multiple", industry "Skincare". The screenshots that follow really walk through the actual creative, which is a woman at a microphone with a scarf and what I would call an expert-podcast aesthetic. The first-frame caption reads "So, antioxidants are super important in skin health", and by 0:12 the captions show "our favorite vitamin C and E", and by 0:21 "antioxidants are what we call free radical scavengers."

This is really the format I wanted to replicate, particularly because the brand isn't sold at all for the first half of the video, and the expert-voice frame really builds trust first. Nuud has equivalent ingredient stories (microsilver, no aluminium), and an equivalent right to be in the educational-explainer lane. I added the idea to the inspiration document, and added a new query ("Importance of antioxidants") to the spider-web sheet for later traffic checking.

The "natural deodorant" query on Creative Center surfaced a higher-CTR ad with a Top 32% CTR, Skincare industry, US region, landing on an Amazon storefront, with the ad caption "Natural deodorant that actually works and looks good". The creative opens with the creator talking through her search for a clean deodorant before actually mentioning any brand, and by 0:13 (which is really half the runtime), she's holding two products and explaining her preference. By 0:29 she's onto specific scents. I would say what made this rank Top 32% wasn't really the product reveal, it was the first 8 seconds, where the personal problem framing really earned the watch-through.

The Fussy | Natural Deodorant ad from the UK was the third Creative Center example I logged in detail. Top 31% CTR, high budget, with the ad caption "The UK's Highest Rated Natural Deodorant!" The two opening frames are split-screen captions: "I have plenty of things on my mind during the day" on the left, and "and the last thing I need to worry about is how I smell!" on the right with the arm raised. Within about five seconds the brand actually gets named ("That's why I choose to use Fussy deodorant!"), and by 0:12 the formulation USPs are already showing ("and without any aluminium or parabens").

I made a specific note next to this ad in the inspiration doc. Mentioning the brand at second 5 really reads as advertising, and I would say that for Nuud's conversion ads that's actually exactly the structure we want, which is fast pain-point and then fast solution. But for our awareness/consideration ads (which is the bulk of what the strategy needs), I would lead with the ingredient or the pain point and delay the brand mention closer to the end of the video. So really it's the same brand, with two fairly different funnel-stage structures.

Probably the single most instructive ad in the entire Creative Center pass was @lorientofficial's "Dark underarms do NOT = bad hygiene", Top 18% CTR, medium budget, Skincare. The retention chart for this video is really what made it stick out for me. The "Remain" line (which is the % of viewers still watching at each second) held at the top 99% of the industry average for the full 1:55 runtime, which is fairly unusual. I would say the hook is doing virtually all the work here, and the format is one we can actually compete on (educational, problem-led, no medical claim, fits Nuud's brand zone fairly exactly).

The next ad I logged wasn't even about deodorant. Averonn's chlorophyll product, US, Beauty & Personal Care, Top 49% CTR, low budget, ad caption "Chlorophyll: The Natural Solution for a Healthier..." The hook the creator opens on is unrelated to the product: "I've never worn deodorant in my life, here's why." That single sentence does enough work that the next 30 seconds, explaining what chlorophyll does, gets a free pass. The audience watches because they've been told there's a surprise coming, then they watch to learn what it is.

The reason this earned a place in the inspiration list: Nuud's actual brand message is semi-shockingly similar. "You don't need deodorant to not smell" is one of the most distinctive messages we can credibly put out, and this ad shows the format that surrounds a message like that. The retention curve confirms it: the chlorophyll ad earned full-watch attention not from product appeal but from a single surprising statement at the front.

Then there's the Kiyome NZ ad, which is only 8 seconds long, with the ad caption "Tired of skincare that does too much? Try Japa..." The video shows the creator looking directly at the camera with the caption "You left your foundation at mine" overlaid on top, with no movement, no product, just the implication. By about second 6 she actually reveals a small jar. The Remain curve on this video is really the cleanest demonstration I have for how a single suspended question can hold attention for the full runtime of a video, and indeed the engagement % by views numbers across the audit kind of confirm that the hook is doing most of the work, with the rest being more delivery than substance.

One more Creative Center ad worth pulling out is Itsneat NZ's "Treating my acne with 100% natural skincare" ad, Top 39% CTR, medium budget. It earned a place in the inspiration list not really because of the topic (we don't cover acne), but more because of the caption choreography. The video runs about 47 seconds, and it uses on-screen text to introduce the problem, name the solution, and explain the active ingredient in really three discrete waves. The captions are reasonably dense (almost too dense in two of the frames), but they actually keep the audience oriented through what would otherwise be a kind of meandering monologue.

I would say the lesson translates fairly directly into Nuud's own video briefs. Awareness/consideration TikTok content for us is often going to be ingredient-focused, which means there is real information to convey, not just a vibe. Without captions the audience drifts, and with too many captions the audience just overloads, and the Itsneat NZ ad really shows the calibration point, which is three to four caption waves per minute, each one a single short claim, paced with the visual reveals.

Outside the Top Ads view, the Trends panel actually surfaces hashtag-level audience demographics. #skincare on TikTok in the 120-day window held a steady 8M posts and 50M overall. The Audience insights showed 53% in the 18–24 bracket, 30% in 25–34, and 17% 35+. That matters because while Nuud's current customer base skews a bit older, our expansion target (the cohort the strategy is really trying to reach) sits in the 18–24 range, and so I would say the audience here is in the right shape. Below the audience chart, the Related Hashtags carousel listed #skincareroutine, #skincaretips, #acne, #skincare101, and #skintok, which are five hashtags I added to the master sheet for traffic checking.

The second hashtag I checked was #skin, which is broader and bigger. 189 posts surfaced for it in the Creative Center, with the related hashtags carousel showing #skincare, #skincareproducts, #facial, #skins, and #acne. The age distribution on this hashtag was actually even more skewed toward the expansion target than the parent: 67% 18–24, 22% 25–34, 11% 35+. So it's tighter to where we actually wanted to grow than the parent #skincare hashtag. I added #skincareproducts to the master sheet for traffic checking, particularly because it was the only one of the five related tags that we didn't already have indexed.

The approach for TikTok was really the same as for Instagram: manual capture of top posts per surviving query, Claude-assisted ingestion of the screenshots into a spreadsheet, and then a second HTML explorer built on top. The header on this one carried the headline counts: 35 unique queries, 133 post observations, 20M total engagement, and 20 posts with view-count data actually exposed.

The "Top 10 queries by post count" view was probably the single most useful page in the whole tool, particularly because it ranked the niches by how much content actually existed in them, which I would say is a cleaner saturation signal than view counts (which the platform doesn't always expose). The chart fairly immediately showed why TikTok needed its own audit, because (related to) "how does sweat work" was the leader at 14 posts captured, followed by "best natural deodorant men" (11), "skincare for men" (11), "best natural deodorant women" (10), "skincare routine" (10), and "aluminum free deodorant men" (10).

Expanding the "how does sweat work" query inside the Explorer shows the ranked post list: Zack D. Films "How Sweating Cools You Down" at 28K engagement, The Infosphere "What is sweat???" at 8.3K, Ancient Health "Why sweating more is actually a good sign" at 5.1K, KonsultaMD "Do you know sweating actually has benefits?" at 2.8K, Meals with Max "Does SWEATING help you to BURN more CALORIES?" at 2.5K, and Kirti Tewari "The liquid that secretes through the sweat glands" at 992. I would say almost every post on this query is really educational, almost every one is presented in "did-you-know" framing, and the highest-engagement video uses a near-identical structure to the antioxidants ad from page 26.

The second-most-engaged TikTok query by post count was "best natural deodorant men" with 11 posts and 2.8M total engagement. The top post, Ethan's "Switch to Based's All Natural Deodorant", registered 1.1M engagement against 928K views and 141K followers, but I flagged it on entry. The headline read as an advert and assumed brand awareness the audience didn't yet have. The posts I actually wanted to learn from on this query were #2 and #3: "switching to a natural deodorant" (ash | low tox · survivor · mama, 991K engagement) and "Are natural deodorants actually good?" (andSomdan, 524K engagement). Same query, much softer framing, real demand.

Expanding "best natural deodorant women" inside the Explorer actually showed an inverted pattern from "how does sweat work". The top post was Rach Lynsey | Beauty "Best natural deodorants I've ever found @Kosas" at 8.0M engagement, 7.9M views, 72K followers. ANITA at 1.2M engagement on "Okay if you're looking for a good smelling…", and alisha at 761K on "natural, non-toxic & hormone balanced". I would say the pattern here is really direct-USP advert framing, and indeed it works. So on the female-coded version of the same query, blunt benefit-led messaging outperformed the soft-education pattern that really dominates the male side.

The eighth query I expanded, "skincare products", surfaced 7 posts and 793K total engagement, with a #1 result that broke the pattern in a useful way: JUSTIN's "Korean skincare products in bio #acne #acnejourn...", 728K engagement, 612K followers. Below it though, the consistently performing format was deinfluencing, Dr Ree's "Deinfluencing you on some viral skincare pro..." at 25K engagement, Dr Nawaz' "The best Skincare products you can get for under £50" at 23K, millie mae's "best skincare for acne" at 6.3K, PrettyWellness' "10 Best Skincare Products (Drugstore + High-End)" at 4.6K.

The deinfluencer angle was the unlock here. The format works because the creator is positioned against a hype cycle, telling the audience what not to buy or do, then naming the alternative. That maps cleanly to Nuud's category position: aluminium-based deodorants are a long-standing hype cycle, and "you shouldn't be putting that on your skin" sits very naturally inside a deinfluencer-shaped video. I added it to the inspiration list as one of the highest-priority TikTok formats for the awareness brief.

The ninth query,"skincare for sensitive skin",had 6 posts and 547K total engagement, but the headline metric of the top post needed flagging on entry. Lukepook's "Glow up Skincare" sat at 340K engagement against just 4.0K followers, an engagement-by-followers ratio of 8,423%. Almost certainly a views-vs-followers field swap in the source data; the engagement-by-followers column on Audit Atlas had a verification flag attached to this row.

The reliable signal in the query was further down. Dr Chris Tomassian's "Dermatologist builds your complete routine" at 14K engagement and 2.0M views, that's a real dermatologist talking, and it ranks. Science sam's "Best skincare routine according to science" at 25K engagement, 69K followers, was the same pattern from a science-creator angle. The brief came out of this clean: pair Nuud with a real dermatologist for at least one anchor TikTok per month. Authority signal earns the watch in a way our brand alone can't yet.

The strategy is new to Nuud, and indeed a new content programme is exactly the kind of bet where leadership is going to expect evidence before scaling spend. So I recommended an 8-week structured test before rolling out at full cadence, which is three new structured pieces per week (one awareness, one consideration, one conversion), running alongside the existing content rather than replacing it. I would say the point of running it in parallel is really that we can test funnel balance without disrupting whatever the current programme is actually doing well.

The slide below is the test recommendation as I framed it to management. The metrics list is per-platform on purpose, with different funnel-stage indicators on each surface, particularly because the audit had already made clear that the two aren't really comparable. To limit risk and cost, the test could optionally be boosted on one smaller regional market first before EU-wide expansion. The objective is explicitly not increased volume, it's more measurable proof that a more balanced funnel improves organic engagement, lowers paid acquisition pressure, and stabilises long-term efficiency.

The whole audit was always going to be fairly useless unless the next person on the brand could actually execute against it without me being in the room. The deliverable that closed the gap between research and production was really a four-page playbook (Instagram awareness, Instagram consideration, TikTok awareness/consideration, plus a general TikTok video formula). Each one combined the hook bank that fell out of the audit with the goals/examples framing that the brand team could brief against directly.

The Instagram awareness page led with the content goals (inform people about armpit health, inform them about the impact of regular deodorant, introduce a "problem and curiosity" angle, encourage people to follow, create interest in natural skincare in general). The example column on the right really ran the hook bank itself: "Sweat does not cause odor" or "Your deodorant might not be solving the real problem", general skincare content linked to armpit health with curiosity-driven angles such as "Why your armpit skin is more sensitive than you think", clear follow prompts framed around value ("Follow for practical armpit health tips most brands do not explain"), ingredient-focused posts that question common deodorant norms, and educational posts on natural skincare benefits.

The TikTok playbook had three pages. The awareness brief really leaned on strong hooks around armpit health, sweating, deodorant use and skincare, with Nuud-specific angles mixed in. The brand-voice instruction was fairly simple: position the account as an expert voice that educates first and sells second. Every video starts with a clear hook (a question, a bold statement, or a surprising fact) to trigger curiosity and improve For You page reach.

The consideration brief shifts the timing of the brand mention earlier in the video, while keeping the educational frame. Hooks like "I always bring this product to the gym" or "How I stayed sweat free in [hot country] this summer without chemicals" really sit Nuud's USP inside a personal-narrative wrapper that still earns the watch. The optional play I added in is replying to customer questions with simple talking videos, particularly because those tend to get fairly high engagement on TikTok at low production effort, and they push users closer to purchase.

One of the constraints I set for myself fairly early on was that this work had to outlive me on the brand, and that really shaped the entire deliverables list. Everything is either a spreadsheet or a small HTML tool that the next person can actually open without my context, and nothing really requires me to explain it. In fact the content philosophy doc reads like an intern handbook on purpose.

For anyone who wants to scroll through the whole research log rather than the page-by-page excerpts above, the complete 101-page document is embedded below. Every screenshot, every annotation, and every "I tried this and it didn't work" detour is present in its original form. The synthesised case study above tells the headline story; this is really the working notes underneath it.

The single biggest cost-saver in this whole project was probably offloading data ingestion to Claude inside structured projects rather than free-form prompts. The pattern below is really the one I documented for whoever inherits the workflow, and I would say it's reproducible on any rebrand, not just this one.

There are three things I want to be honest about before the close. The first is that this really is a topic-discovery and prioritisation exercise, so it tells you what to make, but it doesn't really tell you whether the resulting content will convert at the rates the SMART model is assuming (that's really the 8-week test's job, not the audit's). The second is that this is English-language only, with the follow-up Spanish-language audit scoped but not actually run. And third, the manual platform-audit data is really a snapshot in time, particularly with TikTok where the surface changes pretty fast, so I would say the repeat cadence on this work should really be quarterly rather than annual.

I think it's important to lay all of that out openly, particularly because the audit's value isn't really in any single insight inside it. It's more in being a method that the next person on the brand can re-run from scratch in about two weeks and trust the output of, and that property really survives all three limitations above. The individual numbers don't really have to.