Your Long-Tail Search Results May Tell You What to Write About Next

Happy Wednesday! This week:

Let’s get started!

I’ve been a digital channels strategist for the last 15+ years, and digital channel strategy revolves around one word: content. We create the stories and experiences that connect with our audiences, helping shape what they think, feel, and do.

Sometimes it’s hard to figure out what those audiences care about, and in general, it’s best to lean into the data you have. Typically, we look for signals that reflect a large percentage of our audiences’ interests — which social media posts get a lot of likes, which employee survey results are low, which questions are getting the most upvotes at a town hall, and which topics are coming up on reporter calls. We have to have answers for these. And — specifically for our purposes today — we also need to have content that addresses the most commonly searched terms and phrases that people type into our intranets and external websites.

But what about the smaller ones? We usually ignore them because, frankly, we had to. Thinly-staffed communications teams can’t really dig much deeper — we don’t have the bandwidth to data mine, and even if we did, we’re not resourced to address the edge case results.

I think AI is changing that — particularly when it comes to search traffic.

At the beginning of the AI era — so, about a year ago — I led the rebuild of a company intranet. The goal was simple to state and hard to do: make sure people could actually find what they needed. So we looked at what people were searching for and getting nothing back on. My technology partners gave me a list of all the search terms with 50 or more searches over a 90 day period, and the list was about 500 lines deep. Deduplicating (“parental leave” and “family leave,” for example, are asking effectively the same question) cut the list by at least 50%. My team and I were able to ensure good results for these high-traffic searches within a week after launch.

A few days later, someone complained to me that the intranet wasn’t working for them — “search was broken,” they said. They were a veteran employee with more than a decade at the company, and occasionally booked event spaces for large meetings. About a year prior, though, the company switched vendors for this type of thing, but the old vendor's name had become the office's word for the whole category — the way "Xerox" still means "photocopier" for people who've never touched a Xerox machine. The new intranet didn’t use this term — the content people needed to book a room existed. It just didn't answer to the name half the building was still calling it.

And to this person and a few others, the intranet was, therefore, broken. They kept using the old vendor’s brand. Their searches weren’t voluminous enough to make that term crack that 50+ search threshold, though, so I never saw it. And even if I had, it was among dozens if not hundreds of other esoteric search terms that I could have only addressed if I had an army of interns to help me. That’s why I didn’t ask my tech partners for the full list of search terms — there was simply no way to address all of those rarely-spoken anonymous information requests.

But now, there is. We have an army of interns — it’s called AI.

If I had to do this today, here’s how I’d do it.

First, get the long-tail search list from IT — anything with even one search is fine. Hand it off to an LLM and ask it to give you a readout of what each of those searches mean. If your company gives the LLMs access to corporate data, the chatbots may be able to glean context where you’d otherwise miss it — using my event venue booking example, Copilot may be able to figure out what the term means by looking at OneDrive and SharePoint data, or emails if it has that level of access. And if not, the LLMs typically can access general information at a much larger scale than you or I ever could.

Once you have context for each of those queries, check to see if your intranet (or external website, if that’s where you’re focused) has content that matches the general idea behind the queries. Have the LLMs sort the search terms into three buckets: ones that don’t have any result at all, ones that — in the LLM’s view — yield an incorrect result, and ones that yield a good one. Spot check the second and third bucket — LLMs make mistakes, and you’ll want to adjust your prompting and instructions to adjust for the gaps you find.

Finally, create whatever content is needed. In some cases, you’ll need to take the first crack at it yourself — you don’t want Copilot or any other chatbot making stuff up out of whole cloth. But in many cases, the chatbots can do it for you — the LLM will be appending a line or two to existing content. Whatever gets written from that list gets filed where the topic already lives, never bolted onto the query that surfaced it as some orphaned one-off page nobody will find the next time either.

My plan above is designed for intranets, but the same can be done for your external websites, too — you should be able to follow a similar pattern with similar results. In both cases, before AI, this would have taken forever. With AI, it’s probably a two or three week project, give or take depending on your company’s AI use policies and governance.

And this also unlocks a GEO superpower.

If people are already asking you this question directly — through your own search box — they're also going to ask an LLM the exact same thing. Payoff that search internally, or on your own site, and you've built the actual content that can answer a model too. Fail to payoff the search, and you've got nothing to feed it either — the model will go find someone else's answer, or worse, guess. That's a real upside, and it's worth having. But it's the bonus. The main event is that your search logs are already telling you what your audience cares about and what your executives should be doing thought leadership on. The SEO and GEO lift just comes along for the ride.

You can apply this beyond search — and should.

A similar approach can be used for your other channels. Take every comment and question sent to your company on LinkedIn and Instagram and the like, toss them into a database (a running text document is probably fine — LLMs can handle those), and you’ll have a living record of potential story ideas (or things to watch out for). Internally, you can do the same thing with town hall questions and employee survey verbatims. In these cases, you probably don’t want to publish responses to every single item raised, but you may want to have AI review these, look for trends, and draft responses for those that appear more often that you anticipated. Just in case.

AI's role here is narrower than people want it to be, and that's a feature. It doesn't draft for your readers. It drafts for you — the person doing the triage — turning a spreadsheet of what was previously ignored as “noise” into a ranked list you can act on in an afternoon instead of a quarter.

But today, start with search. Somewhere right now, someone's quietly convinced your intranet — or your website — is broken, over a term you've never seen because it never got loud enough to notice. That’s a problem, but it’s also a content backlog with your name already on it.

Your next steps:

Do this once, and you’ll have a better content offering. Do this regularly, and it’ll feel like you’re actually ahead of your audience.

Claude can be your analyst, editor, and strategist.

But most professionals are using it to fix grammar.

These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.

Sign up for Superhuman AI and get:

Claim your free prompts

Forbes Communications Council ran a piece this week naming a risk that you’ve probably experienced: every competitor's AI tool trained on roughly the same internet, so narrative convergence was baked in before anyone typed a prompt. The piece points to a field experiment from MIT's Sinan Aral and Harang Ju — over 2,200 participants, split into human-only and human-AI teams, each building real marketing ads. The human-AI teams produced about 50% more ads per worker, and by the study's own measure, better copy. But tone, structure, and creative direction all pulled toward the same center — a pattern Aral and Ju are calling "diversity collapse."

I've been making this argument for months and am not alone in doing so — it’s good to get validation, though, because it reinforces that we’ll need to adapt. In an attention economy, our work needs to stand out, and when everyone's using the same assistant on the same brief, being "good" doesn't set you apart anymore.

But I also think — individually — we’re biased toward thinking that our work, AI assisted or otherwise — is going to stand out; whatever magic we sprinkle into our efforts will overcome the diversity collapse. So I think your next step is a gut check.

Pull your last few AI-assisted drafts and a few competitor pieces on the same topic. Ask your friend to review them (and do it yourself, can’t hurt). If they can't tell whose is whose without the byline, that's diversity collapse showing up in your own work.

I spent 15+ years running digital and AI-adoption strategy inside global brands — the kind of enterprise-scale channel work most "AI consultants" have never actually done from inside a company that size. I bring that same playbook to comms teams directly now: workflow audits, executive AI systems, the intranet-and-channel rebuilds I used to run in-house.

A handful of clients at a time. Reply and tell me what you're stuck on.

Looking for a role at the intersection of communications and AI? Here are some opportunities to check out:

Was this forwarded to you by a friend or colleague? Want to get this in your inbox next week? Click here to subscribe!

Quick question: how can I help?

What workflows are you struggling with? Where does AI still feel mysterious or overwhelming? What has worked that you’d like to share with others?

I’m a builder, and I’d love to help you and the rest of The Comms Stack community find great new ways to use AI.

Reply and tell me.
I read every response.

Even a one-sentence reply helps. For example:
“I wish AI could help me with ______.”

Until next Wednesday,

Dan

Advertise | Subscribe