What Is a Good Search Visibility Score? Mine Rose Sixfold and Earned Nine Clicks

The Short Version

There is no good search visibility score, because a visibility score is an index and an index has no absolute meaning. A score of 12 percent is not better or worse than a score of 4,200. They are different scales measuring different baskets.

What there is, though, is a good shape. And the test is simpler than any score: what share of your impressions sit in the top three positions? That number decides whether your visibility is worth anything, and almost nobody reports it.

On this site, that share is 0.85 percent. Which is why a visibility score that rose roughly sixfold in three months has produced nine clicks.

Why I Am Using My Own Site as the Bad Example

I could illustrate this with a client. It would be more flattering and less useful.

The problem with visibility-score advice is that it is almost always written by someone whose score is going up, which makes the score look like the point. I have the opposite situation available: a property where the score went up convincingly, on real demand, in a defensible niche, and the business outcome was almost exactly zero. That is the more instructive case, and I have the warehouse records to show it rather than describe it.

So the numbers below are AISymantix.com, from its first recorded impression through September 15, 2026. Ninety-two days. Every figure comes from Google Search Console query-level data stored in our own warehouse, and you can check the shape of it against the Search Visibility Index we publish monthly.

The 92-Day Receipt

Fourteen thousand three hundred and seventy-one impressions. Three hundred and three distinct queries. Average position 63.3.

Nine clicks.

Here is where those impressions actually landed.

Average position	Impressions	Share of total	Clicks
1	63	0.4%	3
2	40	0.3%	0
3	19	0.1%	1
4 to 10	198	1.4%	5
11 to 20	322	2.2%	0
21 to 50	3,276	22.8%	0
51 and beyond	10,453	72.7%	0

Read the bottom two rows together. 95.5 percent of every impression this site has ever recorded sat at position 21 or worse, and produced no clicks at all. Not a low number of clicks. Zero.

All nine clicks came from the 320 impressions in the top ten, which is 2.2 percent of the total. The other 14,051 impressions were a rounding error with a chart attached.

The Number That Moved, and What It Bought

Here is the part that would look good in a report.

Weekly impressions went from 236 in mid-June to 1,407 by the week of September 7. Average position improved from 88 to 41. Distinct queries went from 25 to more than 90. Every one of those is a real improvement and none of it is manufactured: it is mostly a genuine demand cluster we caught, which I wrote up separately in the Searchmetrics SEO Visibility piece.

Any visibility score calculated on those inputs goes up sharply. Volume rose, position improved, coverage widened. The formula has no choice.

And the click total for that entire improvement was nine.

This is not a failure of the score. The score did exactly what it was built to do. It is a failure of reading the score, because the improvement happened almost entirely inside the range where the click curve is flat and close to zero. Moving from position 88 to position 41 is a genuine gain in every sense except the one that pays. That sentence assumes an average position of 41 still describes something a person could have seen, which is an assumption I take apart later in this article.

Where the Clicks Actually Live

We publish a first-party CTR curve built from 368,474 impressions across 120 properties over a trailing year. The full dataset is here with its sample sizes and caveats, and the formula and worked example are here. For this question, only the top of it matters.

Average position	Impressions	Clicks	CTR
1	1,671	366	21.90%
2	1,079	205	19.00%
3	1,525	90	5.90%
4	2,130	62	2.91%
21 to 50	95,041	177	0.19%
51 and beyond	233,178	70	0.03%

Positions one and two are close to each other: 21.9 and 19.0 percent. Then position three collapses to 5.9 percent.

That is not a gentle decay. It is a cliff between second and third, a factor of roughly 3.2, on samples large enough to take seriously at that end of the curve. Every smooth CTR curve I have seen published has a tidy slope from one through ten. Ours does not. Ours has two good positions, a drop, and then a long stretch of noise.

I want to be careful here, because this is where it would be easy to overclaim. Positions six through ten in our data run 0.80, 1.44, 1.08, 0.92 and 1.16 percent. That sequence is not monotonic, which tells you those individual buckets are sampling noise rather than a real ranking effect. With 25 to 35 clicks per bucket, they should be. I am not going to pretend position seven genuinely outperforms position six.

But the shape survives the noise. Two positions pay well, the third pays a third of that, and past about position five nothing meaningfully pays at all.

So What Is a Good Score?

Given all that, here are the three tests I use instead of asking whether a number is good.

1. What share of your impressions is in the top three?

This is the one that matters most and the one no tool puts on the dashboard. Take your impressions at average position 3.5 or better, divide by total impressions. If that share is rising, your visibility is becoming worth something. If your score is rising and that share is flat, you are accumulating depth, not demand.

Ours is 0.85 percent. That is the honest headline for this site, and it is a considerably more useful sentence than any index value.

2. Is the score rising because of position, or because of coverage?

A visibility score rises when you rank better for the same keywords, and it also rises when you start appearing for more keywords at any depth. Those two are not the same event and they have wildly different value. The second one is what happened to us. Coverage grew from 25 queries to more than 90, and most of the new coverage arrived at position 40 or worse.

Coverage growth is not worthless. It is how you find out what people actually ask, and it is how a young property discovers which questions it can plausibly win. But it should be recorded as research, not as performance.

3. Is the score comparable to its own history?

The moment you change the keyword basket, the score stops being comparable to itself, which was the only thing it was ever good for. Searchmetrics solved this by fixing the basket globally for every domain. If you build your own indicator, you fix yours and then you leave it alone, even when leaving it alone is annoying.

A score you keep improving by adjusting its inputs is not a measurement. It is a mood.

What a Visibility Score Still Cannot Tell You

Everything above treats the score as an honest measurement that was simply read badly. That is fair, but it understates the problem, because the classic visibility calculation measures exactly one surface: ranked organic results.

It cannot see whether an AI Overview answered the question above you. It cannot see whether an answer engine cited you without linking. It cannot see a knowledge panel resolving the query before anyone reaches a result. Our own dataset covers a year in which AI Overview coverage expanded considerably on informational queries, which is one plausible contributor to how thin the curve looks below the top two positions. I want to flag that as a contributing factor rather than a proven cause, because I have not isolated it.

A domain can hold a perfectly stable visibility index while quietly losing the surfaces that now decide whether anyone encounters it. The score will not warn you. It is not built to.

The retrieval half of this is a genuinely different measurement problem, and it is covered in AI Search Visibility vs. Search Visibility and in Search Visibility Index vs. AI Visibility Science.

The Position Number Itself Is Getting Harder to Read

Everything above assumes the word "position" still means what it meant five years ago. I am no longer confident it does, and since the entire visibility calculation rests on mapping a position to an expected click-through rate, that assumption deserves more scrutiny than it usually gets.

Start with what Google actually documents. Compound result features can share a single reported position, so several elements of one block report the same number. An impression can be counted when a result set loads, without the individual result necessarily being scrolled into view. John Mueller has said publicly that the old one-through-ten model is difficult to map onto generative search results in a way that stays useful. None of that is a scandal. It is a reporting system straining to describe a results page that stopped being a numbered list.

The practical consequence is blunt: an average position of 63.3 does not reliably mean anybody scrolled to page six and saw us there. It means the reported average of a metric whose definition now covers several different visual situations.

What Our Own Records Show

We caught a step change across the warehouse in late August. This is every property recorded in our warehouse, the same set of properties in both windows, so composition is held constant even though the absolute totals include some domain and subdomain overlap.

Window	Days	Impressions per day	Clicks per day	CTR	Average position
July 21 to August 17	28	10,079	53.8	0.534%	34.7
August 18 to August 20	3	11,627	69.0	0.593%	35.1
August 21 to September 15	26	13,045	48.0	0.368%	37.2

Impressions per day rose 29 percent. Clicks per day fell 11 percent. Click-through rate dropped 31 percent. More visibility, fewer visits, on the same properties in the same season.

That is the relationship this whole article is about, breaking in public on our own records.

Where I Had the Date Wrong

Google's August 2026 spam update began rolling out on August 18, and my first instinct was to put the break there. Our own data does not support that date. The August 18 to 20 window still behaves like the period before it, with a click-through rate of 0.593 percent, which is actually higher than the four weeks preceding it. The step change is visible from August 21.

That does not clear the update. These rollouts take days to deploy and a three-day lag between announcement and visible effect is unremarkable. But the honest version of the claim is "effects visible from August 21, during a confirmed update that began August 18," not "the break happened on August 18." The mechanism, whether the update broadened query matching or triggered repeated re-evaluation, is a reasonable inference and not something Google has disclosed. You can check the ranking update history yourself.

A Theory I Dropped

For a while I thought the separate generative AI report was the culprit, since Google launched it globally on August 31. That theory does not survive contact with Google's own statement that those AI impressions were already included in the overall Performance report all along. The separate report changed what we could see, not what was being counted. I am leaving the wrong turn in because the correction is the useful part: a reporting change and a measurement change look identical from the outside until you read the documentation.

The Automated-Search Theory, and the Evidence Against It

There is credible vendor research showing that controlled automated Google searches do register as real impressions in Search Console. AccuRanker demonstrated this at a scale of roughly 20,000 searches. It is vendor research, so I weight it moderately, but the underlying point stands: rank trackers and bots can inflate impression counts with nothing human behind them.

It is a tidy explanation, and on this property our own data argues against it. Automated search activity skews heavily to desktop, so if bots were driving the late-August rise, desktop share should have climbed. Here is what actually happened on AISymantix.com across the same two windows.

Device	Impressions before	Impressions after	Share before	Share after
Desktop	4,511	4,835	89.5%	81.2%
Mobile	483	1,110	9.6%	18.6%
Tablet	47	12	0.9%	0.2%

Desktop share fell by eight points while mobile impressions grew 2.3 times. That is the opposite of the automated-traffic signature. So on this property, in this window, I do not think bots are the main story, and I would have happily accepted the bot explanation if the device split had cooperated. It did not.

What I Think Is Actually Going On

My current reading, and I want to be clear it is a reading rather than a finding: a confirmed August update appears to have started serving or testing these pages across a much wider set of query relationships at low relative positions. Modern AI and compound result layouts then make those reported positions genuinely difficult to interpret. Automated searches may add some desktop-heavy noise on top, though not on this property in this window.

I found no reliable evidence that Google formally changed the definition of an impression. The impressions are probably real under Google's own rules. What is no longer defensible is reading them as ordinary people browsing to pages five through eight.

Which loops back to the point of this whole article. If the position input has become abstract, then any visibility index built by weighting position is inheriting that abstraction and reporting it as a clean number. The score gets less trustworthy exactly as the results page gets more complicated, and it will not tell you that it is happening.

The Caveats That Matter

The site was also substantially rebuilt during this measurement window. That matters more than it sounds. When pages are being restructured underneath the impressions, the position data is measuring a moving target, and some of the position improvement from June to September is a different site being evaluated rather than the same site climbing. I would not hand these 92 days to anyone as a clean before-and-after. They are a young property finding its footing while its own structure kept changing.

Our CTR figures come predominantly from niche informational and business-to-business properties. Commercial and brand-heavy sites show materially higher click-through at every position, so if you sell things, your own curve will be kinder than ours. Use the shape, not our magnitudes.

A large share of the impressions in any of these datasets comes from queries the property was never targeting, surfaced briefly at depth and never seen by a human being. That is true of our 233,178 impressions at position 51 and beyond, and it is true of the 10,453 on this site.

Ninety-two days is a short window. Nine clicks is a small number to reason from, and I am not going to build a law out of it. What the window supports is the distribution claim, which is arithmetic rather than inference: 95.5 percent of the impressions were at position 21 or worse, and none of them became a click.

And the causation here runs in only one direction I can defend. Being in the top three is necessary for clicks on these properties. It is plainly not sufficient. Forty impressions at position two earned zero clicks on this site, which is a small sample, and also a reminder that arriving on page one only buys you the chance to be chosen.

What I Am Watching Next

The useful thing about a bad number is that it tells you which question to ask next. Ours is no longer whether the score is improving. It is whether the top-three share is improving.

So the metric on the wall for this property is that 0.85 percent, tracked monthly, published in the Search Visibility Index alongside the rest of the portfolio. If the index value climbs again next month and the top-three share does not move, then we grew coverage and learned something, and we should say that plainly rather than reporting a rising score as a win.

I would rather publish a number that embarrasses us and is real. A visibility score is a fine instrument. It is just measuring the room, not the exit.

If you want help reading your own rankings against your actual pages and offer, request a priced AI Symantix Audit. And if you are here because your old Searchmetrics number disappeared, that story is written up separately, including how the index was calculated and what to use now.