What is worth monitoring
Most competitor monitoring dies within two months because it monitors everything. There are four things that genuinely help you decide:
- Price and promotions. The fastest-changing, fastest-usable data — and the most delicate, see below.
- Messaging. What the headline says, who it speaks to, what it promises. This changes quarterly, not daily.
- Range. New product, discontinued product, new packaging.
- Job ads. The most underrated source: it tells you which way they are developing before they announce it.
What not to monitor: the number of social posts, the follower count, and anything you cannot make a decision from. If you cannot say what you would do differently because of a given line, that line is not needed.
The weekly digest, step by step
1. LIST (once, human)
5-8 competitors. More than that will not fit into one weekly read.
Per competitor 3-5 URLs: price list, blog/news, careers page,
product page. NOT the home page - it changes too often.
2. CHANGE MONITORING (machine)
1-2 checks per page per week is enough. Daily monitoring
gives you noise and costs money.
Output: WHAT CHANGED, not whether something changed.
3. SUMMARY (machine, weekly)
"Here are the texts of the pages that changed this week.
Per competitor, 1-3 sentences: WHAT changed and WHY it
may matter. If the change is cosmetic (a date, a small
wording swap), leave it out. Do not guess at intent,
only flag it when the change fits a pattern."
4. QUARTERLY DEEP DIVE (machine + human)
The same thing over all the changes of 3 months, compared.
This is where you see the direction you miss week by week.
5. DECISION COLUMN (human)
Next to every line: "watching / responding / not interesting".
If for three weeks every line is 'not interesting', you are
monitoring the wrong pages.
It is worth pointing the same system at how your competitors appear in AI answers: on the methodology of that measurement the measuring of visibility post has the details, and there the sample size is the critical point.
What the machine does well, and what it does not
Well: summarising large volumes of text change; recognising a recurring pattern across several weeks of data; skimming foreign-language material quickly; forming categories out of free text.
Badly: the why. If you ask what the competitor's strategy is, you get a confident answer that is a story reverse-engineered from the text. This is the most dangerous output, because it sounds plausible and there is nothing behind it. Which is why it is worth saying it in the prompt: do not guess at intent.
The legal limit: what you may fetch
There are three separate layers, and each one has to be looked at on its own.
Database protection. In a 2021 judgment the Court of Justice of the European Union held that extraction and re-utilisation are prohibited where the maker of the database made a substantial investment in obtaining, verifying or presenting the contents, and the operation poses a real risk to the recoupment of that investment. And it stressed separately: the fact that a database is freely accessible on the internet does not stop it being protected.
Terms of use. According to a 2015 judgment, where a database is under neither copyright nor special protection, the EU database directive does not preclude a contractual restriction on use. In plain terms: the "it is public, so it is free" argument does not hold. The terms are a separate, contractual risk.
Machine-readable objection. Hungarian copyright law knows the text and data mining exception, but the rightholder may reserve their objection in machine-readable form. Once they have, the exception is not available. The European data protection board's guidelines, adopted in July 2026 for public consultation, carry the same logic further: robots.txt, ai.txt, CAPTCHA and a login wall carry significant weight in the balancing of interests.
The practical translation: monitoring the public text of public pages is generally fine; bulk extraction of structured data, content behind a login and circumventing a technical objection are not. The same boundary applies to collecting reviews, which the sources of customer words post covers in more detail.
The competition law limit: this is where it gets genuinely interesting
This is the part most articles keep quiet about, even though in price monitoring this is the main risk.
The leading case is a 2016 EU judgment. The operator of a booking system sent a message to the agencies using the system saying that the discounts available through it were capped. According to the Court, the companies that were aware of the message were presumed to have taken part in the concerted practice — unless they publicly distanced themselves, reported it to the authority, or rebutted the presumption with other evidence, for instance by showing that they systematically applied discounts above the cap.
The key word is becoming aware. The price monitoring software is not the infringement. The infringement is learning about a competitor's pricing constraint through a shared system or channel and not distancing yourself from it.
There is an example of the software becoming part of the facts. In 2018 the European Commission fined four electronics manufacturers for resale price maintenance, 111 million euros in total. There the price monitoring software did not figure as an infringement in itself but as context in the facts: it made it possible to identify quickly the retailers deviating from the price, so the manufacturers could concentrate on a few of them, knowing that the correction would run across the market fast through the reaction of the software.
And enforcement is speeding up right now. In March 2026 the British competition authority opened an investigation in the hotel sector over algorithm-assisted information exchange; the German authority acted against a platform's algorithmic price control mechanism; the Dutch and the Italian authority are examining airline dynamic pricing. We found no Hungarian competition supervision case about algorithmic pricing — but following a government decree from late 2025, where the product that is the subject of a competition supervision procedure contains an AI system, a separate specialist authority joins the procedure as well.
The practical rule: monitor public prices, and decide independently. What to avoid: exchanging pricing intentions through a shared system or an industry channel; setting up automated price following in a way that effectively ties your price to a competitor's; and any agreement at all about a price ceiling or a price floor.
What it costs
Orders of magnitude, based on list prices:
- Page change monitoring: from $0 a month (a few pages, hourly frequency) to around $350 (several hundred pages, for a team).
- Price monitoring for e-commerce: typically between $100 and $400 a month, depending on the number of products and the refresh frequency. The list price of the more serious packages is often not published at all.
- Keyword and visibility monitoring: around $100–500 a month, cheaper with annual payment.
- Competitive intelligence platform: no list price, on request.
The cost of the model that writes the summary is negligible next to that. The money goes on the monitoring, not the summarising — which is why one or two checks a week matters so much.
Common misconceptions
- "It has to be monitored daily." Most useful changes measure themselves in weeks. Daily monitoring gives you noise and costs money.
- "What is public is free to take." Terms of use are enforceable as a contract, and a freely accessible database can still be protected.
- "Using price monitoring software is lawful, so there is no risk." The software in itself is not an infringement — but becoming aware, and the response to it, can be.
- "The machine will tell me the competitor's strategy." It will not. It tells you a reverse-engineered story, in a confident voice.
- "The more competitors, the better." Above five to eight nobody reads the report, and that is how the system turns into a dead process.
What to take away
Competitor monitoring works when it looks at few pages, rarely, and every line has a decision attached to it. The machine takes over the summarising and the pattern recognition — working out intent must not be handed to it. And the two limits are not technical ones: the terms of use decide what you may fetch, and competition law decides what you may do with what you found out.