Skip to content
BrandWater AI
Strategy

The competitive intelligence you can get for free by prompting AI platforms

By BrandWater AI Research · Published 1 November 2026

1 November 2026 17 min read

Quick answer

AI assistants are a free, repeatable lens on how your category is perceived: which competitors get named together, what language consistently attaches to each one, and which attributes come up unprompted. Run the same prompts several times each, by hand, before investing in any tool, and treat a single answer as an anecdote rather than a finding.

There is a version of competitive research that costs nothing but time: open ChatGPT, Gemini, Perplexity and Claude in separate tabs, type the exact questions your customers already ask when they are choosing between you and a competitor, and read what comes back. Done carelessly, this produces nothing but a handful of screenshots and a strong opinion based on one lucky or unlucky answer. Done with a bit of discipline, it produces a genuinely useful, free first pass at how your category is perceived, before you spend anything on a tool to formalise it.

Why this works as a research method at all

AI assistants answer category questions by drawing on a mix of training knowledge and, when search is enabled, retrieved web content, including the comparison articles, review sites and forum threads your actual customers read before they talk to you. In effect, asking an AI assistant "what's the best X for Y" is asking it to summarise the same public conversation a real prospect would encounter if they did their own research. That is genuinely useful: it's a compressed view of your category's public reputation, assembled from sources you may not otherwise read closely.

It is not, however, a stable or authoritative source in the way a market research report is. The same question can return a different list, in a different order, with different language, from one run to the next, because the answer is sampled, not looked up. Treating a single run as the finding is the single most common mistake people make with this kind of informal research, and it's an easy one to avoid once you know to watch for it.

What to actually look for

Rather than reading answers for whether you "won" or "lost", which tempts you to overreact to one good or bad run, look for patterns across several runs on the same prompt.

Who gets named alongside you, consistently

If the same two or three competitors show up next to you across most runs of the same prompt, that's your real competitive set in this category, as perceived by AI assistants, which may not be the same as the competitive set in your internal strategy deck. It's worth noticing when a company you don't think of as a direct competitor shows up repeatedly, that's a signal about how the category is actually being framed in the sources these answers draw from.

What language consistently attaches to each brand

Pay attention to the adjectives and framing, not just whether a brand is named. "Known for enterprise support," "the budget option," "popular with freelancers" are doing real positioning work, and if that framing doesn't match how you'd describe yourself, that's a specific, actionable gap between your intended positioning and your actual public perception, whether or not it's fair.

Which attributes come up unprompted

If you ask a neutral "what should I look for when choosing X" question and the same three or four criteria come up across runs, pricing transparency, ease of setup, customer support responsiveness, whatever they are for your category, that tells you what the sources being drawn from think actually matters to buyers. If your own marketing emphasises different criteria, that's a mismatch worth investigating.

The sampling discipline this requires

A single answer to a single prompt on a single platform is one draw from a range of possible answers, exactly the same caution BrandWater applies to its own measurement. The practical version of this discipline, done by hand rather than with a tool, looks like running the same prompt three to five times, in a fresh conversation each time so earlier context doesn't bias later answers, across each platform you care about, and writing down what's consistent versus what only showed up once. If a competitor appears in one of five runs, that's noise. If they appear in four of five, that's a pattern worth taking seriously.

It's worth testing a handful of different phrasings of the same underlying question, too, not just repeating the identical prompt. "Best accounting software for a small agency," "accounting software for a 20-person agency," and "what accounting tool do agencies use" are all getting at roughly the same intent, but can surface different answers depending on how a platform's search step interprets the wording. If your brand only appears under one specific phrasing and not the others, that's useful information about how narrow your current visibility actually is.

Building a lightweight version of this before you invest in tooling

You don't need a spreadsheet template or a formal process to start. A genuinely useful first pass looks like: pick five to eight real questions your customers ask when evaluating your category, drawn from sales call notes, support tickets, or your own "why did you choose us" customer interviews rather than invented from scratch. Run each one three times on each of the two or three platforms your customers are most likely to actually use. Note, for each run, which brands appear, in what order, and with what framing. After a week of doing this, you'll have a rough but genuine picture of your current competitive position in AI answers, built entirely by hand.

A simple manual tracking sheet
PromptPlatformRunBrands namedYour position & framing
Best accounting software for a small agencyChatGPT1 of 3Illustrative: Brand A, Brand B, (you)Named third, described as "budget-friendly"
Best accounting software for a small agencyGemini1 of 3Illustrative: Brand A, (you), Brand CNamed second, no specific framing given
Accounting software for a 20-person agencyChatGPT1 of 3Illustrative: Brand A, Brand BNot named

This table is illustrative

The brand names and positions above are placeholders to show the format. Fill it in with your own real prompts, your own platforms, and what you actually observe, run by run.

Where this informal method breaks down

Manual checking has real limits, and it's worth being honest about them rather than pretending a spreadsheet of hand-run prompts is a complete picture. It doesn't scale past a handful of prompts before the time cost becomes prohibitive. It's easy to unconsciously run more checks when you suspect a problem and fewer when things look fine, which biases what you notice over time. It gives you a one-time snapshot, not a trend, so you can't easily tell whether something is getting better or worse without redoing the whole exercise from scratch at a later date, with no guarantee you ran it exactly the same way the first time. And it's manual-labour-intensive to extend across your full competitive set, every platform, and every category question that actually matters to your business, rather than the handful you had time to check.

That's the specific gap a structured, repeated audit closes: the same discipline, running a defined prompt set across platforms on a schedule, with every run logged and a sample size attached to every rate, done automatically instead of by hand, is exactly what turns an occasional spot-check into a trend you can actually act on.

What to do with what you find

If a competitor consistently shows up with stronger framing, look at what public content is likely feeding that, do they have more detailed comparison pages, more recent reviews, clearer documentation. If your category's unprompted criteria don't match your marketing emphasis, that's worth a conversation with whoever owns your messaging, not necessarily a content overhaul, but at least an acknowledgment of the gap. If you're simply not named at all for prompts your customers plausibly ask, that's the clearest, most actionable signal: start by checking whether you have any page that directly and clearly answers that exact question, because a surprising number of visibility gaps turn out to be a content gap first, not a ranking or reputation problem.

Reading tone and sentiment, not just presence

It's tempting to reduce this whole exercise to a binary: were you named or not. The more useful read pays attention to the tone an answer takes toward each brand it mentions, which is a layer of information a simple presence count misses entirely. Two brands can both be named in the same answer, one described in neutral, factual terms ("offers a free tier with basic reporting") and the other in warmer, more confident language ("widely regarded as the most reliable option for small teams"), and that difference in tone is itself a competitive signal worth tracking, even though both brands technically "appeared."

In practice this means reading the actual sentence structure around each brand mention, not just scanning for whether the name is present. Watch for hedging language attached to one brand and not another ("some users report," "it's worth checking whether"), for which brand gets the first mention versus a later, more parenthetical one, and for whether a brand is described with a specific, concrete benefit or a vague, generic one. None of this requires special tooling to notice by hand, it just requires reading the full answer carefully rather than keyword-scanning it for your brand name and moving on.

Briefing content and PR teams with what you find

The informal competitive check described in this piece produces exactly the kind of concrete, specific observation that's genuinely useful to hand to a content or PR team, far more useful than a vague "we should improve our AI visibility" directive. "Across five runs of this exact question, on these two platforms, competitor X was named with this specific framing and we weren't named at all" gives a content team something they can actually act on: a gap to fill, a specific comparison to address, a specific claim to potentially substantiate or counter in their own published material.

The framing matters when you deliver this internally. Presenting it as "here's what we observed, across N runs, on these specific prompts, on this date" reads as a grounded, credible input to planning. Presenting it as "AI thinks we're worse than competitor X" invites exactly the kind of dismissive pushback ("it's just a chatbot, who cares") that stalls any follow-up action. The discipline and the framing are doing the same job: making this look, and actually be, like a real, repeatable observation rather than an anecdote dressed up as a finding.

A note on doing this within the platforms' terms of use

Manually typing prompts into a consumer AI assistant's chat interface and reading the answers is ordinary use of the product, the same as any other user asking questions. Where this can get complicated is if you try to scale it up through automated scripts or scraping against a platform's consumer interface rather than its sanctioned API, which is where most platforms' terms of service draw a clear line. If you ever want to move from occasional manual checks to something more systematic and automated, that's exactly the point where it's worth using a tool or method built on sanctioned access rather than improvising automation against a consumer chat interface, both for reliability and to stay clearly within the platform's terms.

Watching for your own content reflected back at you

One specific pattern worth watching for during these checks: an AI assistant describing your own brand using language that traces, almost word for word, back to your own marketing copy, rather than to any independent evaluation. This isn't necessarily a problem, it often just means your own site is a well-trusted, frequently retrieved source for questions about yourself, which is a reasonable thing to be. But it's worth distinguishing from a competitor being described using language drawn from independent reviews, comparison sites, or forum discussion, because that's a meaningfully different, often stronger signal, independent sources corroborating a specific claim carry more weight than a brand simply describing itself favourably, and noticing which kind of sourcing is behind each brand's description gives you a more honest read of the relative strength of each position.

This distinction matters practically too. If your own strong points are only ever being repeated from your own copy, with no independent corroboration anywhere, that's a gap worth addressing through genuine third-party coverage, reviews, case studies published by actual customers, press mentions, rather than more confident language on your own pages, which an AI assistant, and a skeptical reader, can reasonably discount as self-description rather than independent evidence.

Doing this across a team without duplicating effort

Once more than one person on a team starts running these informal checks, it's worth a small amount of coordination to avoid the two most common failure modes: several people independently checking the same handful of obvious prompts while a longer list of equally important ones never gets checked by anyone, and two people reaching quietly different conclusions about the same prompt because they ran it on different platforms, or at different times, without realizing it. A short shared list, who's checking which prompts, on which platforms, and roughly how often, solves most of this without needing any formal tooling, it's simply a matter of making the division of labour visible rather than assuming it'll sort itself out organically.

It's also worth designating, even informally, one person as the point of continuity for this effort, someone who holds the running log described earlier in this piece and who can answer "what have we found so far, and when did we last check" without the team having to reconstruct that history from scattered memory or old chat messages. That continuity is what separates a team that's done a handful of interesting one-off checks from a team that's actually building an accumulating, genuinely useful picture of how its category is perceived over time.

Checking this against your actual win-loss data

The informal AI-assistant check described in this piece becomes considerably more useful when you cross-reference it against whatever win-loss or customer-feedback data you already have, rather than treating it as a freestanding exercise. If your sales team's win-loss notes already show prospects citing a specific competitor advantage, and the same advantage shows up consistently in how an AI assistant frames that competitor, that's a corroborated, higher-confidence finding, worth prioritizing over something that only showed up in one source or the other. Conversely, if an AI assistant consistently frames a competitor favourably for a reason your own win-loss data never mentions, that's worth a second look too, it might be a perception gap that hasn't yet reached your sales conversations, which can be either an early warning or a sign the AI framing doesn't actually reflect what matters to real buyers in practice.

How this compares to traditional competitive intelligence

Traditional competitive intelligence work, win-loss interviews, mystery shopping, analyst briefings, review mining, has always tried to answer a version of the same question this exercise answers: how is our category actually being talked about, by people who aren't us. What's different about checking AI assistants is speed and cost. A win-loss interview program takes weeks to schedule and run; a round of AI-assistant checks across your core prompts takes an afternoon. That doesn't make it a replacement for deeper qualitative research, a five-run prompt check will never surface the nuanced reasoning behind a specific lost deal the way a real customer interview can, but it does make it a uniquely fast, cheap first signal, worth running far more often than a formal competitive intelligence program ever could be, precisely because it's cheap enough to repeat.

What this exercise can't tell you about why

Even a carefully run, well-repeated version of this check has a specific, important limit worth stating plainly: it tells you what's being said, not definitively why. If a competitor is consistently named with stronger framing, the AI-assistant check itself doesn't tell you whether that's because they have genuinely better product documentation, more favourable press coverage, a larger volume of positive reviews, or simply a more recent comparison article that happened to rank well in whatever search step fed the answer. Those are different underlying causes with different fixes, and distinguishing between them requires the follow-up investigative work described in the companion piece on diagnosing why a competitor gets recommended instead of you, not just repeating the same prompt check a few more times.

It's worth treating the manual competitive check described in this piece as the first half of a two-part process: notice the pattern here, cheaply and quickly, then diagnose its specific cause using a more structured follow-up before committing real effort to a fix. Skipping that second half and jumping straight from "we noticed a gap" to "we're publishing more content" is exactly the shortcut that leads to effort spent on the wrong remedy.

A note on seasonality in your own checks

Category perception, as reflected in AI answers, isn't necessarily stable across the year, especially for businesses with any seasonal pattern to their own demand. A check run once, in isolation, during an unusually quiet or unusually busy period for your category, may not generalize to how things look the rest of the year, and it's worth being aware of this rather than treating any single check, however carefully run, as a permanent, year-round baseline. Where it's practical, running the same core prompt set at more than one point in the year, and noting whether the competitive picture shifts, gives a more complete, more honest view than a single snapshot ever can, and helps you avoid over-reacting to a pattern that turns out to be tied to a specific season rather than a durable, year-round competitive reality.

Knowing when to graduate to something more structured

Manual checking earns its keep as a starting point, but there's a fairly predictable point at which its limits start costing you more than a structured approach would. That point usually arrives when you notice yourself wanting to compare this month's answers to last month's and realising you didn't keep careful enough notes to do that reliably, when the number of prompts you actually care about has grown past what you can reasonably re-run by hand on a regular cadence, or when more than one person on your team is doing informal checks independently, with no shared record, duplicating effort and reaching inconsistent conclusions. None of those are failures of the manual method, they're just signs it's served its purpose as a cheap first look and the next useful step is tracking this with the same rigor you'd apply to any other metric that drives real decisions.

What a negative result actually tells you

It's worth spending a moment on the case where this entire exercise turns up nothing notable at all, your brand and your main competitors all show up with roughly even frequency and broadly similar, neutral framing, across every prompt and platform you checked. That outcome is easy to read as a wasted exercise, since there's no dramatic gap to act on, but it's actually a genuinely useful finding in its own right. It tells you that, as far as AI-mediated discovery is concerned, your category currently looks close to a level playing field, which is valuable information whether or not it's exciting, because it means any future shift away from that even footing, in either direction, is something worth noticing quickly rather than assuming is just normal variation.

A flat, even result is also a reasonable prompt to widen the check rather than conclude there's nothing more to learn. If the obvious, high-level category prompts show no gap, it's worth trying a few more specific, use-case-driven variations, the kind of narrower questions a more qualified prospect might actually ask, since competitive differences often show up at that more specific level even when the broad category question looks evenly matched across the board.

Doing this well

  • Run the same prompt several times before drawing any conclusion from it.
  • Test a few different phrasings of the same underlying question, not just one.
  • Track framing and language, not just whether you were named.
  • Use real customer questions, not questions you'd like to be asked.
  • Know the limits: this is a snapshot, not a trend, until you repeat it on a schedule.
  • Read tone and sentiment around each mention, not just whether the name appears.

Frequently asked questions

How many times should I run the same prompt before trusting the result?

Three to five runs, in a fresh conversation each time, is a reasonable minimum for an informal check. If a brand or framing shows up consistently across that many runs, it's a real pattern. If it only shows up once, treat it as noise.

Does it matter which AI platform I check first?

Check the platforms your actual customers are most likely to use, which may not be the one you personally default to. If you don't know, checking the two or three largest general-purpose assistants plus whichever platform is most relevant to your industry is a reasonable starting set.

Is this kind of manual check a substitute for a proper AI visibility audit?

It's a useful, free first pass, not a substitute. It doesn't scale, it's a one-time snapshot rather than a tracked trend, and it's vulnerable to your own biased attention. It's a good way to decide whether the problem is worth investing in measuring properly.

Is it worth keeping a permanent archive of past answers, not just a summary of what was consistent?

For your most important category prompts, yes. An archived answer from a past check can be useful later for confirming exactly when a particular framing or competitor mention first appeared, rather than relying on memory of roughly when it was noticed.

What if I get completely different answers every time I run the same prompt?

Some variation between runs is normal and expected, that's the nature of a sampled answer rather than a lookup. What matters is whether there's a consistent pattern underneath the variation, which is why running it multiple times rather than once is the whole point.

Should I try these prompts while logged into my own accounts or logged out?

Try both if you can, and note any difference. Some platforms personalise answers based on account history or location, and a logged-in result may not reflect what a new, unknown prospect actually sees.

My competitor appeared in every run with much better framing than us. What's the first thing to check?

Look at what's likely being cited to support that framing, their comparison pages, recent reviews, documentation, before assuming it's an unfixable reputation gap. Often it traces back to a specific, improvable piece of public content rather than something structural.

Should we worry if a competitor seems to dominate every single prompt we try?

Before concluding that, check whether you're testing a narrow, repetitive set of prompts that all happen to favour the same framing, or a genuinely varied set covering different phrasings, use cases and customer segments. A competitor that dominates one narrow angle of your category looks very different from one that dominates every angle you can think of, and it's worth testing a genuinely broad set before treating either conclusion as settled.

Is it worth checking what AI assistants say about our own brand name directly, not just category questions?

Yes, it's a useful complementary check, since it can reveal whether an assistant has any outdated or inaccurate information about your own company specifically, separate from how you're positioned against competitors on a category question. Both checks are worth running, as they tend to surface different kinds of gaps.

Cite this page

BrandWater AI Research. (2026, 1 November 2026). The competitive intelligence you can get for free by prompting AI platforms. https://brandwaterai.in/blog/free-competitive-intelligence-from-ai-platforms

How we work

Figures are dated and linked to their sources. Where none exist we say so. Read our methodology and AI transparency pages.

Put the theory to work on your brand.

Start with a first audit. We show where you appear, who is named instead, and what to fix first.