If your only competitive intelligence process is checking which blog topics a rival has covered that you haven't, you are missing most of what actually moves a client's decisions. AI competitive intelligence gets more useful once you point it at pricing pages, messaging changes, product launches and the answers AI engines give when a buyer asks about your client's category. This guide is for agency delivery leads who need a repeatable process they can run across a whole client roster, not a one-off research sprint for a single account. By the end you will have a seven-step method for using AI tools for competitor research on pricing, positioning and product changes, plus a client-ready way to report what you find.
This guide covers everything besides topic coverage: what a competitor charges, what they're saying, what they just shipped, and what AI itself tells a buyer comparing them to your client. One thing it will not do is repeat content gap analysis, the process of finding topics competitors cover that you don't. That already has its own workflow, so we won't retread it here.
Step 1: Define the decision and shortlist who you're watching
This starts with knowing why you're watching, not just who. Start by asking which decisions this actually needs to inform: a pricing page update, a sales objection your client's reps keep hearing, a response to a competitor launch, a differentiation line for the homepage, or an AI-visibility report. Then build a short list of the direct alternatives your client actually loses deals to, the ones that come up in real sales conversations. Add an adjacent competitor only if it would change a real decision. For each one, record the product, buyer segment, geography and official domain, and assign an owner on your team along with a review date.
Before you turn on any tool, open a separate intelligence register for each client. A shared spreadsheet across accounts is how one client's pricing notes end up informing another client's strategy by accident, and that's a mistake you don't get to make twice.
You'll know this step is done when every competitor on the list has a reason for being there and someone is accountable for watching them. Two names in the industry can share a brand, or a competitor can sell three products under one domain, so resolve that ambiguity now rather than after you've built a monitor around the wrong page.
Common mistake: tracking a dozen competitors because it feels thorough. A bigger watch list means more noise, more of your monitoring budget spent on brands nobody at the client cares about, and less confidence in the two or three comparisons that actually matter. Cut the list before you build anything on top of it.
Your register should hold: the client and workspace, competitor and product, category, buyer persona, country and currency, source page, the date and time you observed something, the captured text or screenshot, price and billing unit, the specific claim, what type of change it was, the old and new value, source type, whether it's verified, your confidence level, how significant it is, who should own the response, and the next review date. That looks like a lot of columns for a first pass, but most rows will only need a handful filled in. This register is the backbone of AI competitive intelligence work across a roster: without it, every account starts from zero.
Step 2: Establish a dated baseline for pricing and packaging
Once you know who you're watching, capture their pricing page as it stands today, along with any feature comparison, FAQ or terms page that explains what's actually included. Build one row per plan, not one row per company, because a monthly number on its own tells you almost nothing.
Record the monthly price and the annual price separately, and note whether the "annual" figure shown is the actual yearly charge or just the monthly-equivalent rate once you divide it out. Capture included seats, usage limits, minimum commitments, gated features, trial terms, and whether the top tier just says "contact sales." Before you tell a client that a rival is cheaper, make sure you're comparing the same seat count, the same required features and the same billing term. Save a dated copy of the page, because pricing pages change without notice and you'll want the receipt.
When a plan has no public number, write "not publicly listed, quote required" rather than guessing. If a prospect shares a quote with your client's sales team, or your client hears a number secondhand, log it with its own date and source, clearly separate from the public list price. Do not publish someone's half-remembered number as if it's the competitor's standard rate.
If you're using AI to pull this together, feed it the captured page text and ask for one row per plan: displayed amount, currency, billing interval, charge unit, seat or usage allowance, feature gates, trial and quote-only status. Tell it to write "not stated" for anything missing rather than filling in a guess, and to keep the exact phrase and date behind every populated cell. Then check the output against the actual page before anyone sees it. AI is fast at structuring this kind of text; it isn't a substitute for you looking at the source.
Pro tip: a colleague on your team should be able to reproduce your pricing comparison from the underlying terms alone, without you in the room to explain it. If they can't, the row needs more detail, not a footnote.
Step 3: Capture messaging and positioning across every touchpoint
Pricing tells a client what a competitor charges. Messaging tells them what a competitor is trying to convince a buyer of, which is often the more useful signal for a client deciding how to differentiate.
Baseline the exact wording: the homepage headline and subhead, the product page's core promise, the audience or industry labels a competitor uses, their proof points and differentiation claims, comparison page copy, and the main call to action. Alongside the site, check the ad creative a buyer might actually see. Meta Ad Library lets you search by keyword or advertiser once you set a location and category, and it's genuinely useful for seeing what a competitor is running right now. Treat it as evidence that an ad exists and what it says, not as evidence of how well it's performing or who it's targeting.
Feed dated before-and-after excerpts to AI and ask it to group the changes by buyer, problem, benefit, proof and call to action. Ask it to return the exact phrases that were removed and added, classify each change, and label anything that looks like a strategic shift as a hypothesis rather than a fact, along with what else you'd need to check to confirm it.
You're done when you have a messaging matrix showing each claim word for word, where and when it appeared, who it's aimed at, and a clearly separate note on anything that might be a positioning shift. One edited headline is not a strategic pivot on its own. Before you tell a client a competitor has repositioned, check whether the same language shows up again across product pages, announcements and ads, not just in one spot you happened to catch.
Common mistake: lifting a competitor's claim into your own client's marketing because it sounded good. What a competitor says about itself is a data point about their strategy, not a fact you get to borrow.
Step 4: Put your key pages on a monitoring watchlist
Put the pricing table, hero text, product pages, changelog and any relevant help docs on an explicit watchlist first. Manually rechecking all of it by hand doesn't scale past two or three clients, so this is the step where AI-powered competitor monitoring earns its keep. Pick a monitoring method for each page based on what you're actually watching.
For visual, page-region change detection, a tool like Visualping works by having you enter the page address, select the exact region you care about, write a plain "alert me when" condition, choose a check frequency, and set where the alert should go. It can send you a highlighted screenshot, a text difference and a short AI summary when something changes, which is useful for catching a price move or a new plan appearing without you staring at the page yourself. Free tiers on tools like this usually cap how often they check and how many pages you can watch, so read the plan limits before you promise a client hourly monitoring on a free account. That's true across most vendors in this space, not just this one.
Not every page needs the same monitoring method. For a repeated, structured table like a full pricing grid, a scraping and monitoring tool such as Browse AI can train a robot to capture specific fields, then run that capture on a schedule and flag additions, changes and removals between runs. It's a better fit than a screenshot tool when you need the same structured row every time, like plan name, price and seat count, rather than a general sense that "something changed."
A third layer catches what the other two miss. A broader tool like Google Alerts, which emails you new search results matching a topic, is worth adding as a catch-all for launches or mentions you wouldn't think to watch for directly. It's not a substitute for a precise comparison of a page you already know matters. It catches what you didn't know to look for.
You're done with this step once every high-priority page has an owner, a captured baseline, a monitoring method, a sensible check interval, and a confirmed test alert that actually reached someone. Good AI tools for competitor research at this stage are the ones that let you narrow the watched area precisely, not the ones that promise to watch everything. That distinction matters more than any feature list.
Common mistake: monitoring an entire dynamic page instead of the specific region that matters. Cookie banners, rotating promos and personalized sections create constant false alarms. Narrow the watched area to the pricing table or the specific text block, and check whether what you're seeing depends on your login state, region or browser before you report it as a real change. A reasonable starting cadence is daily or twice-weekly checks on a high-impact pricing page, weekly checks on messaging and release notes, and a full monthly baseline refresh, with quieter clients checked less often. Treat that as your own starting point, not an industry standard, and adjust it to what each account actually needs.
Step 5: Verify product claims before you act on them
Every feature change your monitor catches needs a second look before it goes anywhere near a client deliverable. This kind of launch-response work is exactly where AI agents for competitive research earn their keep, pulling the initial comparison together fast so a person can spend their time checking it instead of building it. Compare the announcement or changelog entry against the official release notes, help documentation, and the actual product or plan page, since these can describe different stages of the same rollout.
Note whether what you're looking at is a genuinely new capability, an expanded limit, a renamed feature, something removed, or a feature gate that moved between plans. Ask AI to produce a difference table between the old and new state, then have a person on your team check that table against the actual source text before it's treated as fact. If a capability has only been announced and isn't live yet, or the rollout is limited to certain accounts, say exactly that rather than implying every customer already has it.
You'll know this is done when your record states what changed, which product and which customers it affects if it's live, its current availability status, and what's genuinely still unknown. A marketing headline announcing a feature is not proof that every customer can use it, and even help documentation can lag behind what's actually shipped. A changelog entry is a reason to check further, not a reason to close the loop.
Step 6: Check what AI itself is telling buyers about the competition
Everything so far has been about what competitors say on their own pages. This step is about what AI engines say when a buyer asks them directly, which is a different question and one that page monitoring can't answer on its own.
Pick buyer questions that put real alternatives side by side, either a direct comparison of two named products for a specific requirement, or an unbranded question about options in that category. Capture the full answer along with the date, the engine, which brands got mentioned, which sources got cited, any price or product claims the answer made, and where that answer disagrees with what you know from the competitor's actual pages. Repeat the same set of questions on a fixed schedule, because a single answer on a single day tells you almost nothing about a pattern.
This is where a platform like DeepSmith fits into the process. Agencies can define or select the buyer prompts worth tracking, then read mention rate and citation rate separately, see the full answer history, compare share of voice against named competitors, and see exactly which competitor pages got cited for each tracked prompt. Discover Prompts can suggest a starter set of questions based on a client's product and audience, which saves the blank-page problem of guessing what buyers actually type. Use this for the specific job of tracking how a client and their competitors show up in AI answers, not as a stand-in for watching a rival's live pricing table or their internal product roadmap, which it was never built to do. What engines a workspace can track depends on the plan, so don't promise a client coverage the account doesn't actually have.
This step closes with three concrete things you can point to, not a feeling. You're done here once the client can see which prompts and engines produced each observation, which competitor pages got cited, and whether something an AI answer got wrong is flagged as exactly that, an AI-answer observation, rather than presented as a verified market fact. Nothing here should still be a guess.
Common mistake: treating a citation in an AI answer as proof that the engine endorses that product, or treating one changed answer as evidence of a lasting shift in positioning. The set of prompts you track is a sample of the questions buyers actually ask, not every question that exists.
Step 7: Triage the changes and ship a client-ready brief
By this point you likely have more observations than any client wants to read. This step is where AI can draft the summary, but a person on your team has to approve every implication before it goes out.
A useful brief for an agency client holds: the decision it's meant to inform, the verified change, the old and new state side by side, the exact source and the date you captured it, which buyer segment it affects, your confidence level and any open questions, a recommended action, and who owns it. Route confirmed, consequential pricing or product changes to the account lead and sales enablement right away. Save repeated messaging shifts for a strategy review. Send anything cosmetic to a weekly digest instead of an individual alert. Keep the full evidence register behind the shorter client-facing summary, so anyone can trace a claim back to its source if they need to. That way the summary stays short without losing the trail.
You're done when every item on your list has either been assigned an action with evidence attached, queued for further verification, or dismissed with a stated reason, and the next review date is set.
Common mistake: forwarding the raw alert feed to a client, or presenting an AI-generated guess about a competitor's motive as if it were settled fact. "Their listed annual allowance changed" is something you observed. "They're panicking about churn" is a story, and it's not yours to publish. If you're building comparative marketing off any of this, the FTC's policy on comparative advertising permits honest, non-deceptive comparisons, but it doesn't make an unverified claim about a rival safe just because you found it through AI. Escalate the things that are both verified and tied to an actual buyer decision, and let everything else sit in the register until it earns a reaction.
What to do next
This is the step most agencies skip once they're busy. Keep each client's competitor list, captured evidence, tracked prompts, alert destinations and approval owners separate, even when two clients compete in a similar space. Give every account the same record structure and review checklist, so a new team member can pick up any client's register and understand it immediately, but never let one client's pricing assumptions or sales notes bleed into another's brief.
Set a recurring review date for each account, even a light one, so this doesn't quietly stop after the first month. If you're already tracking how a client shows up in AI answers, DeepSmith's AI Visibility module gives you the prompt tracking, citation data and competitor share of voice that step six depends on, alongside the content production side of the platform for turning what you learn into on-brand pages. You can try it with your own client's data on a free trial before committing to anything.
