Key findings
2 firms · Jan 2025 to Nov 2026
Time to lose half the gain: 3 months Both firms had lost half of what they gained by month 3 after stopping.
Stop AEO and half the visibility you built is gone within three months. It doesn’t fall to zero, but firms that keep going pull away.
54%
below peak today
250 ranking keywords in Nov 2026, against a combined peak of 5481
2.7×
still above the pre-program level
Neither firm fell back to its starting point: 250 keywords now vs 92 before
373 vs 38 and 74
firms that kept going, vs the two that stopped
Median visibility index of four firms that kept their programs, Nov 2026 (Nov 2025 to Jan 2026 = 100)2
Visibility built through answer engine optimization behaves less like an asset and more like a position. It is held against engines that keep re-ranking and competitors that keep publishing. When a firm ends its program, how quickly does that position erode, and how much survives? We asked our own portfolio data.
Method#
Metric. The Perfectus visibility index counts the US organic keywords a site ranks for each month, from Perfectus Labs visibility tracking. It is a modeled estimate of search visibility. It is not traffic, calls or revenue, and we keep those measures separate.
Indexing. The sites differ in size by an order of magnitude, so each is indexed to its own November 2025 to January 2026 average (=100) and compared on a log scale, where equal distances mean equal percentage changes.
Control group. Four tax-resolution firms in the same category that kept their programs running through 2026. They are matched on vertical and size, not randomly assigned.
Decay model. For each stopped firm we fit the months from the stop through November 2026 two ways: a simple exponential, by least squares on log visibility, and an exponential decay toward a floor. We compare the two by AIC. Intervals come from the standard error of the slope and from a bootstrap over months. The simple-exponential decay rates are 7.4% a month for the Texas firm (95% CI 3.9 to 10.7%) and 7.7% for the Georgia firm (95% CI 3.8 to 11.4%).
Contrast. The main estimate is a difference in differences: each stopped firm's change in log visibility from baseline to July to November 2026, minus the same change for the control median. A log point gap of 0.69 is a factor of two.
Robustness. We repeat the contrast with two alternative baselines, a later comparison window, and without the control group's strongest grower, and we check where each stopped firm sits among all six.
Finding 1: the Texas firm decayed to a floor#
In January 2025 the firm ranked for 19 keywords. By January 2026 it ranked for 200. The program ended at the end of February, and the decline was already under way that month as the work wound down.
The shape is the finding. A simple exponential describes the post-stop months only moderately well (R² 0.68). A decay toward a floor fits clearly better (R² 0.87, AIC lower by 7). The site covered half the distance to that floor in about a month, then settled near 68 keywords, roughly a third of peak. Of the 181 keywords gained during the program, half were gone by March, one month after the stop.
Dots are monthly counts and lines are fitted decay curves. Modeled estimates of search visibility, not traffic, leads or revenue.
Finding 2: the Georgia firm declined later and more slowly#
First-party analytics during the program show the build: monthly search impressions rose from about 5,500 in February 2025 to about 27,900 in March 2026, and clicks from 64 to 112. Ranking keywords peaked at 348 in April 2026. The program ended in mid May.
Here the simple exponential is the better description (R² 0.74). The floor model does not improve AIC. The fitted rate is 7.7% a month, a half-life on total visibility of about 8.7 months, with a bootstrap 90% interval of 5.6 to 12.4. Half of the 275 keywords gained since February 2025 were gone by August, three months after the stop. The last four months sit between 188 and 199, which looks like a floor forming, but seven points cannot separate a floor from continued slow decay.
Fitted to the same simple model, the two firms lose ground at almost the same rate, 7.4% and 7.7% a month. What differs is how front-loaded the loss is.
Finding 3: the category moved the other way#
The obvious alternative explanation is the engines themselves. Google ran spam updates in June, in August and from late September into October 2026. The control group rose through all of them.
From baseline to July to November 2026, the control median rose 1.11 log points, roughly a tripling. The Texas firm fell 0.78 log points, down 54%, and the Georgia firm fell 0.24, down 21%. The differences in differences are 1.89 and 1.34 log points. By the end of the window, the two firms stood at about 15% and 26% of where they would be had they tracked the control median.
Two stopped firms against four active. A pattern, not a causal estimate.
The gap holds across specifications. Alternative baselines of September to November 2025 and October 2025 to January 2026 give gaps of 1.57 to 1.80 log points for Texas and 1.38 to 1.61 for Georgia. A later comparison window, September to November 2026, widens them to 2.00 and 1.47. Removing the strongest control grower, which went from 18 to 108 keywords in 2026, cuts the gaps to 1.22 and 0.68. Both stay clearly positive.
The distribution check is the honest constraint. One control firm fell 0.79 log points, as far as the Texas firm. Both stopped firms sit below the control group's interquartile range, ranking second and third lowest of six, but not outside its full range. A rank permutation test on six firms gives p of about 0.27.
Limits#
Sample. Two stopped firms and four controls, all small sites. At this scale monthly counts move in steps, and single months are noise.
Measurement. The index is modeled visibility, not observed traffic, leads or revenue. First-party analytics for the Georgia firm end with the engagement.
Identification. Firms chose to stop. They were not assigned, and whatever drove that choice may also affect visibility. We have no view of site changes after an engagement ends, and the control group is matched by category, not by prior trend.
Model fit. Decay curves fitted to seven and ten monthly points carry wide intervals, as the bootstrap ranges show. We hold the shape with more confidence than the exact rates.
What it means for firms considering a pause#
Two things are consistent across both firms. The program-built gain reverses quickly, with half of it gone within one to three months. And it does not go to zero: by November the Texas firm held about 30% of its peak and the Georgia firm about 55%.
If budget is the constraint, a reduced program generally costs less than rebuilding after a gap. If you do stop, set a baseline first: ranking keywords, search impressions and clicks, and the AI answers you appear in today.
FAQ#
What is a visibility half-life?#
The number of months after a program ends until half of the visibility gained during it is gone. In this data it was one month for the Texas firm and three months for the Georgia firm.
Did a search algorithm update cause the decline?#
Not on its own, as far as this data shows. The control group rose through the same update months, and the gap survives alternative baselines, a later window and the removal of the strongest control firm.
Does visibility fall to zero after AEO stops?#
No. Both firms kept a meaningful share of their peak. The Texas firm settled near a fitted floor of about a third of peak.
Does lower visibility mean fewer leads?#
Not necessarily, and this note does not measure leads. Ranking keywords measure search visibility. Traffic, calls and signed cases are separate measures, and each needs its own tracking.
To see where your firm's visibility stands before you decide, Book an AEO Audit. We will show you what the engines say about you today and what is holding that position up.
