Algeria AgencyDigital marketing
What you are really paying for: the search terms report
You choose keywords; you are billed on real searches. The gap between the two is the whole job, and it is read once a week.
A search account runs on a useful misunderstanding worth naming once: you choose keywords, and you are billed on searches people actually typed. They are not the same list, and the gap between them is most of what happens in an account.
That gap is visible. It sits in one screen the platform supplies — the search terms report — which the advertisers paying for it almost never open. An account can be well built and spend for months on searches nobody has ever read.
This article is the maintenance work: the weekly reading, the exclusion list that becomes an asset, the structure that keeps that list usable, relevance, automated bidding and the point at which it has something to work with, knowing which word made the phone ring, the hours, the competitor buying your name, and the budget spent by midday.
The article that accompanies this one asks the prior question — do these searches exist, and in what quantity — along with the three languages, the landing page and the payment constraint. It is worth reading first: it decides whether to open an account, this one decides what you do with it afterwards.
You buy words, you pay for searches
The distinction is mechanical. A keyword is a purchase instruction; a search term is what somebody actually wrote. The platform matches the two with a latitude it decides, and that latitude is wider than most advertisers imagine.
The result is that an account which bought "air conditioner repair" also pays for "car air conditioner repair", "cheap air conditioner", "how to repair an air conditioner yourself" and "price of a new air conditioner". Three of those four lead nowhere for a repairer, and they are billed exactly like the first.
This is not a defect to be corrected, it is how the system normally works, and it has a reason: nobody can list in advance every phrasing of one intention. The latitude exists to catch the ones you had not thought of, and it catches everything else as well.
The practical consequence is therefore that the work is not choosing better keywords at the start. It is reading what was billed, regularly, and removing what should not have been. It is subtraction, and it is what separates a maintained account from a launched one.
It also changes how to judge an account at a glance. Do not ask for the keyword list — it is always presentable. Ask for the exclusion list and the date of the last search terms reading: those two facts describe the real account.
The search terms report, once a week
The reading takes about twenty minutes and is always done the same way. Open the search terms for the last seven days, sort by descending spend, and go down the list asking one question at each line: can this person buy something from me?
There are three answers and each produces an action. Yes, and the term is already covered: nothing to do. Yes, but it is a phrasing you had not anticipated: add it as a keyword, because a term bought directly generally costs less than the same one caught by latitude. No: exclude it.
Sorting by spend is what makes the exercise sustainable. A search terms list contains hundreds of lines most of which cost a few dinars, and reading them all is a task nobody does two weeks running. The top of the list concentrates the money; twenty lines is almost always enough.
The frequency is not negotiable at the start and relaxes later. Weekly for the first two months, because that is where most of the exclusions are built; then fortnightly, then monthly when each pass yields two or three new lines.
Finally, record what you exclude, with the date, in a document outside the platform. That list is what you take with you if you change account, agency or platform, and it is the only part of the work that cannot be redone: keywords are recovered in an hour, two years of exclusions are not.
One engine, therefore one list to maintain
Advertisers are regularly offered a duplicate of their account on a second engine, on the grounds that being present costs little. The spend is indeed not the problem, and that is precisely why the argument is badly framed.
What a second account costs is the weekly pass described above: a second reading of the terms, a second exclusion list built from nothing, a second structure to keep coherent. The scarce resource in a small account is not money, it is the regular attention of the person holding it.
And searches made in Algeria are concentrated to a degree that makes the trade obvious. In July 2026 one engine accounted for 96.96% of searches and the second for 1.92%. Doubling the maintenance load to reach under two per cent of searches means halving the quality of the work on the remaining ninety-seven.
The practical conclusion is not "other engines are useless" but "they are not worth your weekly pass". If one day your main account is maintained, stable, and time is left over, the question comes back; while the terms reading is not weekly on the first, opening the second is a choice against yourself.
The same reasoning applies to any duplication inside one platform — a second account for a second brand, a second structure for a test. Every object created adds a pass to make, and an unread account does not do less damage than one that does not exist: it does more, since it spends.
Statcounter Global Stats, Algeria, July 2026
The exclusion list is an asset that compounds
An exclusion does not wear out. A term removed today stays removed, and it goes on saving money every month for years without anybody touching it. It is the only part of an advertising account with that property: everything else has to be redone.
That makes an account’s progress curve unusual and often misread. The first weeks produce many exclusions and a clear improvement; then the rate of discovery slows, and people conclude the work has stopped producing. In reality the effect of the exclusions already made continues, silently.
Three families cover almost everything, and it is useful to hold them in mind for reading quickly. Information searches, which want to know rather than buy — "how", "why", "what is". Free-of-charge searches, which exclude buying by definition. And job searches, which almost every service account pays for without knowing.
A fourth family belongs to this market: searches aimed at another wilaya. A tradesman in Algiers regularly pays for "plumber Oran", because geographic targeting applies to where the person is and not to what they write. That is a word exclusion, not a zone setting.
Structuring so the exclusions stay usable
An exclusion applies at a place in the account, and the choice of place determines whether it protects everything or only a corner. Placed too low it has to be repeated elsewhere; placed too high it cuts off part of the account where the term was legitimate.
The rule that avoids both is simple: what is never relevant to the business goes as high as possible, what is irrelevant to one service goes on that service. "Job" and "free" are the first kind and should be set once and for all; "rental" is the second if you both sell and rent.
It also decides the size of the structure. An account cut into twenty groups requires each exclusion to be thought through twenty times, and whoever holds it ends up placing them at random. Few groups, broad, with high exclusions, get maintained; many narrow groups do not, whatever the good will.
One technical point that is expensive when ignored: a broad exclusion can cancel a keyword you bought elsewhere in the account, without warning. After setting a general exclusion, check that your important terms still receive impressions the following week — a thirty-second check that avoids a month of unexplained silence.
Relevance: what moves it, and what does not
The platform scores the match between what somebody searched, what your ad says, and what the landing page contains. That score affects what you pay and whether you appear, which makes it a cost lever as much as a visibility one.
What moves it is ordinary and fits in one sentence: the same phrasing in all three places. If the search says "air conditioner repair Algiers", the ad should contain those words and so should the landing page. This is not subtle optimisation, it is consistency, and most accounts fail at the third step.
What does not move it: raising the bid, changing a button colour, adding more keywords, or rewriting the ad to make it more persuasive. Many advertisers spend their time on those four actions and conclude the system is arbitrary.
The landing page is worth repeating because it is the step nobody wants to take: sending every ad to the home page is the commonest shortcut and it degrades the score of every group at once. One page per service, however plain, beats one beautiful home page for everything.
Finally, relevance is an indicator rather than an objective. An account can have excellent scores and produce no calls, and a mediocre account can support a business. Look at it when the cost rises without explanation; do not pursue it for its own sake.
Automated bidding, and when it has something to work with
Automated strategies promise to manage bids better than a person, and they often do — on one condition that is rarely stated: they need enough recorded conversions to learn anything. Below that they optimise on noise.
The threshold is not expressed in dinars but in conversions per month, and the platform states it itself. An account producing eight calls a month has nothing to feed an automated strategy with, whatever its budget, and manual bidding does the job better there.
There is a precondition many advertisers skip: the conversion has to be recorded. An account where the call arrives on a mobile phone with nothing reporting it has no conversions as far as the system is concerned, and handing it the bidding means asking it to optimise towards an objective it cannot see.
The sensible sequence is therefore always the same: record conversions first, stay manual long enough to accumulate them, move to automated, and allow two to three weeks before judging. Changing strategy restarts a learning period, as on feed platforms, and judging it after three days always produces the same wrong conclusion.
Knowing which word made the phone ring
Here the conversion is very often a call, and a call leaves no trace in the advertising account by default. The advertiser sees clicks and an invoice, hears the phone ring, and has no way of connecting the two.
The consequence is worse than an inconvenience: without that link, every decision in the account is taken on cost per click, the only figure available. So the expensive keywords get cut — often the ones that make the phone ring — and the cheap ones are kept, cheap because nobody wants them.
Two solutions exist depending on means. The simple one is a separate number used only in the ads, whose calls are logged apart — not precise to the keyword, but it already separates advertising from everything else. The complete one is the platform’s call tracking, which attaches the call to the search term.
There is also a method that costs nothing and almost nobody applies: asking. "How did you find us?" put to every caller for two weeks produces a crude but real figure, and it is often enough to correct the first error, which is believing the advertising returns nothing because nobody watches it return anything.
The worst case is the account that records a tap on the number displayed on mobile and counts it as a call. That is not a conversion but an intention: the person may not have pressed, may have hung up, or may have found the line busy. An account run on that figure systematically overstates what it produces.
Hours and days follow whoever answers
The delivery schedule is treated as a setting and it is an operational decision. If the conversion is a call, delivering outside the hours when somebody picks up means paying for ringing in an empty room.
The first version is therefore crude and correct: deliver during opening hours, plus a short margin either side. The margin exists because people search before they call and because a form or a message can be left out of hours, but it stays a margin, not a second working day.
Only then look at the report by hour and by day. What you find there is rarely what you expected: Friday often produces searches without calls, the evening produces enquiries that call back the next day, and some hours cost more for an identical result.
One local case is worth noting because it is frequent: businesses that receive most of their calls after the working day, between six and ten in the evening. A schedule copied from office hours deprives them of their best window, and nobody notices because the account spends normally during the day.
Somebody is bidding on your name
One day somebody searches for your business by name and sees a competitor’s ad first. This is legal in most cases — buying a word is not misappropriating a mark as long as the ad itself does not use your name — and it is a common practice.
The first question is whether you are actually losing anything. Somebody searching your name knows you, and they move down one line. The harm is real but limited, and it is measurable: check whether searches on your name have lost clicks since the ad appeared.
The usual answer is to bid on your own name, and it is less justified than people say: you pay for a click you would have had for nothing, unless the organic result is weak or the competitor’s ad genuinely sits above it. Check before spending, by searching your own name from a device that does not know you.
What is almost always justified, on the other hand, is dealing with the content of the competitor’s ad. If it uses your name in its text, that is a different subject from advertising and it is settled by a report to the platform and then, if necessary, on trade mark ground — see the article on brand protection.
We do not do it in the other direction, and that is a position rather than legal caution: buying a competitor’s name produces mediocre clicks, opens an escalation the small advertiser always loses, and turns around the day the competitor answers with a larger budget.
The budget spent by midday
An account whose daily budget is consumed before midday is not working at half strength: it is working on the half of the day that is not necessarily the right one. It is a symptom, and it has three causes requiring three different answers.
The first is a term too broad that absorbs everything — a case the weekly search terms reading settles with one exclusion. The second is a budget genuinely insufficient for existing demand, which is good news badly presented. The third is delivering at every hour while demand is concentrated.
Do not raise the budget before ruling out the first cause, because a larger budget given to an overly broad term produces exactly the same problem an hour later, and more expensively. The order is: read the terms, exclude, narrow the schedule if needed, and only then decide whether the budget should rise.
When the cause really is the second, the conclusion is the one people like to hear and should still verify: demand exceeds the budget, so each additional dinar finds a taker. Verify it on the real cost per call rather than on lost impression share, which always rises and always sells an increase.
What we do, and what we refuse
What we do is bounded. We make the weekly pass on the search terms, we keep the exclusion list in a dated document that belongs to you, we put conversion recording in place before any automated strategy, and we align the ad and the landing page on the phrasing actually searched.
We refuse to hold a search account without that weekly reading, including when the engagement is presented as a simple setup. An unread account does not stay as it is: it drifts towards whatever the platform’s latitude decides, and it spends while it drifts. A setup without maintenance is work we know in advance will degrade.
We also refuse to bid on a competitor’s name, for the reasons given in section 10 — the quality of the clicks, the escalation, and the reprisal — and we say so before being asked, because it is a frequent request.
And we do not promise a position or a cost per click. The first is not for sale and the second depends on what others are bidding at the same moment. What we commit to is the reading, the list and the measurement. There is also work that belongs to you and weighs more than the account: answering, and calling back the people who got no reply.
Frequently asked questions
How often should the search terms report be read?
Weekly for the first two months, then fortnightly, then monthly once each pass yields only two or three new lines. Twenty minutes is enough if you sort by descending spend and stop at the bottom of the first twenty lines: that is where the money is, and reading the hundreds of lines below it is what makes people abandon the exercise.
Why are we paying for searches we did not buy?
Because a keyword is a purchase instruction and a search term is what somebody actually typed; the platform matches them with a latitude it decides. That is not a defect: it exists to catch phrasings you had not thought of, and it catches the rest too. The work is removal, not better choosing at the start.
Should we also be present on a second search engine?
Rarely, and not for a budget reason. A second account doubles the maintenance load — a second terms reading, a second exclusion list — to reach 1.92% of searches made in Algeria against 96.96% for the first. While the weekly reading is not being kept up on the main account, opening the second degrades both.
Which exclusions should be set first?
Four families cover the essentials: information searches, free-of-charge searches, job searches, and searches aimed at another wilaya — the last because geographic targeting applies to where the person is, not to what they write. Put what is never relevant at the highest level of the account, once.
Can we let the platform manage the bidding?
Yes, once two conditions are met: conversions are genuinely recorded, and there are enough each month for the system to learn — a threshold it states itself. An account with eight monthly calls, or one where the call is reported nowhere, is asking an automation to optimise towards an objective it cannot see.
A competitor is buying our name. Should we respond?
Measure first: check whether searches on your name have lost clicks since, and search for yourself from a device that does not know you to see what actually appears. Bidding on your own name makes you pay for clicks often already won. If the ad uses your name in its text, however, that is a trade mark matter and it is reported.
Where we come in
Twenty lines sorted by spend are enough the first time, and the result always surprises. The following week, nobody opens the report again.
- We reread the report every Monday and send you the exclusions added.
- We date every exclusion, because the list is the part that cannot be rebuilt.
- We give the document back if we stop, in its original format.
If your account spends under twenty thousand dinars a month, keep this list yourself: it takes ten minutes and does not usefully delegate.
Read next
Google ads: does the intent exist, and in what quantity
This platform does not create demand, it answers a question already typed. Somebody still has to type it.What advertising budget to set — and how to know
There is no published rate card for Algeria. Here is what the available data actually lets you conclude.A live campaign: the moves of the first week
The budget is set and the money is going out. What to look at, what to change, and above all what not to touch yet.
Let us talk about your project
A free audit, no commitment: we look at your online presence and tell you what is holding it back.