In San Francisco the people buying bookkeeping, dental work and kitchen remodels are engineers, and they behave like engineers. They read four pages before they dial. That habit changes which figures in a search report are worth acting on.
Getting at the numbers is trivial; a verified property produces more of them than a six-person practice will consume. The difficulty is separating three things that look identical in a table: a buyer you persuaded, a buyer who was only counted, and a client company that no longer exists at that address.
Your customers are technical. Your practice is not.
The distinguishing fact about selling professional services here is not who the competitors are. It is who the customers are. A tax preparer in the Financial District, an orthodontist on West Portal, a general contractor in the Sunset — each sells to people who spend their working lives evaluating systems. They compare on spreadsheets, open six tabs, and want to finish as much of the transaction as possible before speaking to anyone. The call is the expensive step, not the first one.
That raises the floor on what a page must contain. A firm publishing hourly rates, a scope table and a plain answer to "what does this cost" beats one whose services page ends in an invitation to get in touch; to this audience that invitation reads as a refusal. Meanwhile the owner is a dentist or a carpenter with no background in any of it, judged on documentation by people who write documentation for a living.
- The visit is not the decision. A technical buyer arrives, reads, leaves, and returns through a different door two days later. Several sessions separate the first arrival from the signed client.
- Unanswered questions disqualify you. A missing price, a vague scope, an unstated turnaround — each sends the reader to the next tab instead of your contact form.
- Self-service is assumed. Buyers expect the site to let them qualify themselves. Anything requiring a conversation to learn reads as concealed.
- The owner is not the audience. Pages that please the owner — warm, general, reassuring — fail the reader who wanted a number.
One record belongs to you; the other describes everybody else
Two families of screens sit side by side and appear to report the same thing. They do not, and treating a disagreement as an error is how an afternoon disappears.
Search Console replays what Google's serving systems wrote down about a domain you verified: every occasion a result of yours went in front of somebody, and whether that somebody took it. Within that fence the record is authoritative. Beyond the fence it says nothing at all. When a competing practice publishes the pricing page you never got around to writing, nothing announces it — your line bends downward and you supply a story.
Rank tracking comes at the problem from outside. Because it queries the results page under controlled conditions, it can list who else occupies the slots, score those domains for authority, and total the terms held in common with each. Past the click it goes blind. Eight screens sit on the first side of that line and six on the second, and the argument for keeping both under a single login is that neither can be made to answer the other's questions.
What your own property recorded
The observed record: real people, real queries, real choices, delayed by two days.
- Dashboard across verified sites. Takers, appearances, ratio, trend lines and health flags for every property in the account at once.
- Keywords, pages, devices, countries, traffic. Five cuts of one log, each answering something the others cannot.
- Keyword dynamics. Which terms entered or left the top three, ten and thirty — movement rather than a snapshot.
What the results page looks like from outside
The external reading: who else is there, how strong they are, and what you share with them.
- Ranking score and average placement. A portfolio reading across every domain in the account, with a twenty-eight-day trend curve.
- Competitors in your term space. Rival domains with Domain Authority and a count of keywords you hold in common.
- Top keywords and best pages. Terms by search volume against your placement, and the pages carrying most of them.
| The question in the room | Which record answers it | Why the other one cannot |
|---|---|---|
| Did anyone choose us? | The click column in your own log | Being shown is not being chosen |
| Who else is on that page? | Competitor listings with authority scores | A rival never shows up inside your own property |
| What do buyers actually type? | The query breakdown | Volume figures elsewhere are modeled estimates |
| Is this term worth pursuing? | The volume-sorted keyword list | An impression count measures exposure, not appetite |
| Was it us, or the whole results page? | Reading the pair side by side | Either alone yields a confident wrong answer |
Each of the four top-line figures misleads on its own
Any report opens with the same quartet: how often you were shown, how often you were taken, the ratio between those, and where you sat on average. All four definitions are sound, and all four in isolation point at the wrong work — by four different mechanisms, which is why they are worth separating.
Impressions
The tally of times a result of yours was drawn onto a screen somebody looked at. Trivially inflated, and routinely mistaken for interest.
- Grows when irrelevant phrasings start matching you
- Grows when one page spreads across a broader term set
- A fall here beside flat clicks is normally welcome news
Clicks
The one figure tied to a human preferring you to everything else on screen, and the one most affected by the wording of your listing.
- Carries no information about the visit itself
- Piles up in a handful of rows and nowhere else
- At low volumes it lurches from week to week
Click-through rate
A fraction, so either half can move it. Going up may mean a sharper title, or it may mean broad exposure quietly disappeared.
- Almost worthless as a site-wide figure
- Swings wildly depending on placement
- Judge a row against its own history, not a benchmark
Average position
Your best placement per appearance, then averaged over all of them — every phrasing, every location, every screen size at once.
- One obscure phrase down at sixty ruins the mean
- Gets better when weak terms drop out entirely
- Useless until you narrow to one named phrase
Together the quartet behaves. More appearances without more takers points to phrasings that were never going to convert. More takers on flat appearances means the wording improved. A stable placement with a decaying ratio nearly always means a firm above you sharpened its listing — or published the price you still have not.
The click that matters is rarely the first one
One log, two arrangements. Sorted by phrase it records what the market wanted; sorted by URL, what you produced. Owners reach for the second first, because those rows correspond to work they remember commissioning — which is precisely why it uncovers so little. A list of your own documents cannot contain the thing you never wrote.
Here that split does something specific. A researching buyer does not make one search and one click. They land on an explanatory article, leave, return through a comparison phrase, read a service description, come back a third time by typing the firm's name, and only then use the contact form. Your report holds all of it as unrelated rows, with nothing joining them into one person.
- Plenty of appearances, almost no takers. Phrases where you are shown and passed over. Occasionally the listing is at fault; more often the page lacks whatever the phrase promised.
- Rising traffic on your own name. People typing the firm's name are returning, not discovering. That row is the tail of a path that began elsewhere in the report.
- Explanatory pages with clicks and no conversions. Often your most valuable pages. They open the process; something else closes it. Judging them on their own conversions retires the wrong page.
- Query rows with no page behind them. Demand you serve by accident. The single most productive hour available in this report.
Then run the two against each other on purpose. Choose a phrase that matters and see which URL was served for it. The answer is often not the one you would have nominated, and that mismatch is an unmade content decision neither arrangement surfaces alone.
Where the reader sat, and what they were holding
Two breakdowns are consistently underused, and here both carry more than usual. The country split matters because a practice on Montgomery Street does not serve a local market in any tidy sense: its clients are firms whose finance lead sits elsewhere and whose office manager searches from another address entirely.
The device split matters differently. One firm often has two audiences on two kinds of hardware — the buyer researching from a laptop mid-afternoon, and the operations person calling a plumber from a phone at eight in the morning. They need different pages, and this is the only view where that shows.
Both render as heatmaps rather than tables, which sounds cosmetic and is not: an outlier becomes obvious without reading a single number. Countries you never marketed to, or a mobile ratio half the desktop figure on one phrase, show up in a second. The reporting layer, including the country and device heatmaps, exports to CSV or JSON up to ten thousand rows.
Last year's baseline may describe customers who are gone
Date comparison is the feature everyone uses and the likeliest source of a confident wrong conclusion. Standard practice sets the current window against the same window a year earlier. In a stable market that is sound. This is not a stable market.
Companies here close, merge, downsize and relocate on a cycle few cities match. A client roster can turn over substantially inside eighteen months with nothing going wrong at your firm, and the phrasings turn over with it, because they came from a population of businesses that has partly been replaced. A year-on-year fall in something like "startup bookkeeping" may say nothing about your website and everything about how many startups remain.
| Comparison | What it answers well | Where it breaks here |
|---|---|---|
| 28 days versus previous 28 | Did a recent change do anything? | Seasonal noise, holidays, one loud week |
| 90 days versus previous 90 | Is the direction real? | Slow to react; a bad month hides inside it |
| Year on year, totals | Almost nothing, on its own | Assumes the same customer base existed |
| Keyword dynamics, rolling | What entered and left the top ten | Movement without volume can flatter you |
The substitute is to stop comparing totals and compare named sets. Take the twenty phrasings that produce enquiries and track those against themselves. If nineteen held and one collapsed, that is a page problem worth an afternoon. If all twenty softened evenly while new phrasings appeared underneath, the market moved and the site did not fail. The keyword dynamics view, which records entries into and exits from the top three, ten and thirty, beats any totals comparison here, because it reports movement between states rather than the height of a bar.
What the query list is quietly telling you to publish
Reports improve nothing. What improves things is the page you write after reading one, and here the phrase list asks for one page over and over: the one with the number you have been avoiding.
Look for phrasings built around cost, rates, hourly, how much, pricing, fee. They arrive with respectable exposure and a ratio that embarrasses the rest of the table: you are shown, and the reader picks somebody whose listing promises an answer. For this audience a rate card with honest ranges and stated assumptions beats a beautiful page ending in a form.
The number, with its conditions
A range, plus the variables that move it inside that range. This reader does not need certainty; they need to know what determines the figure.
- State what is included and what is not
- Give the turnaround in days
- Say what makes a job cost more
One page per real question
When a single URL is catching a dozen unrelated phrasings, what you have is two or three documents that were never pulled apart.
- Filter the page view to one URL
- Read every phrase it collects
- Split where the intents diverge
The generative research screens push the same way: intent classification across your phrase set, pages the model flags as leverage for expansion or linking, and a content-gap reading against the competitors sharing your terms. Useful for narrowing, still not a decision.
What practice owners ask after the first look
Our position improved and traffic went down. What happened?
Usually one mechanism: a set of broad, weak phrasings that used to place you near the bottom of page four stopped matching you. They dragged the mean down while contributing almost nothing, so their departure flatters the number and removes reach at the same time. Find the vanished phrasings and see how much exposure went with them before calling it good or bad.
Our tracker reports three and the console reports eight. Which is wrong?
Neither is wrong. One is a single controlled sample taken the same way each time; the other blends every genuine appearance, in whatever city, on whatever device, with whatever personalization applied. A gap that stays wide is normally a map: strong at the point where the sample is taken, thinner across the rest of the territory your clients search from.
How do we tell which page brought in a client who read six of them?
It cannot: the log records phrases and clicks, not people or sessions. Ask on the enquiry form how they first came across you and read that beside the phrase list. Expect an explanatory page at the start of the path and your firm's name typed directly at the end.
Should we compare against last year at all?
Only at the level of named phrases, and only where the clients behind them still exist. Compare totals across a year here and you measure how many companies stayed as much as how the site performed. Where a client category has thinned out, say so rather than letting the chart imply a marketing failure.
How soon after a change should we expect to see something?
Treat the tail end of any window as provisional — the last two days have not finished arriving. After that, allow four to eight weeks before movement means anything at all. Anyone drawing a conclusion at the ten-day mark is reading noise, and for a practice with modest click volumes that noise is loud.
The point where the numbers hand the problem back
All of this describes; none of it decides. You learn which phrasings delivered people, where they landed, how the shape changes by hardware and geography, and how this stretch sits against the last. Nothing establishes whether the client behind a phrasing was one you wanted, whether the missing page is realistic in the week you actually have, or whether the practice above you holds that slot deliberately or by inertia.
Those are judgments, and where readers evaluate better than you present, judgment is the whole game. Where the record is thin — a client category that has largely left, a service launched last quarter — the data narrows very little, and the honest move is to say so rather than over-read a small table. More of our material on that pass sits in the articles section.
Budget an afternoon for the first read. Set the range to ninety days, order the phrase list so heavily shown and rarely taken rows come first, and ignore the totals at the top of the screen. Then reorder by takers and hunt for anything phrased as a question. The first list shows where you are skipped; the second, what you turned out to be good at without intending to.
To see both records assembled rather than collected screen by screen, connect a verified property and open the dashboard — analytics, rank tracking and exports run off one account, with campaign automation from $149 per month per domain and a managed tier at $500. What usually surfaces in a practice here is not a ranking problem but a list of things people asked and nobody answered plainly. Which you take on depends on capacity, and capacity is not in the report; our engagement outlines cover having somebody else run the pass, and the report builder and export tools make the result something a client will read.