The restaurant marketing metrics that predict cover count

Cover count is a lagging number. By the time it drops you have lost the month, and you are looking at drift from several weeks earlier.

What an operator wants is the equivalent of a prep check at three o'clock. Something that says Friday will be a problem while it is still Wednesday.

Which numbers actually move before covers do?

Four, and they are all free to look at.

Direction requests and calls from your Business Profile. The closest thing to intent you can measure. Somebody asked their phone how to get to your restaurant. Watch the trend across months, not the number in a week.

Menu page visits. The menu is the highest intent page on a restaurant website. People who read a menu are deciding. If menu visits fall while everything else holds, something changed in how findable that page is, and the structure of that page decides a lot of it, which is covered in how to build a menu page Google can read.

Review pace against the restaurants above you. Not your star average, which barely moves. How many new reviews arrived this month compared to how many arrived for the profiles ranking ahead of you. Whitespark's 2026 Local Search Ranking Factors survey, in which 47 local search experts scored 187 factors, puts recency of reviews at number 11 and a sustained influx of reviews rather than bursts at number 14. Darren Shaw puts the standard plainly: "The moment you stop getting new reviews, you're going to see your local rankings start to slip."

Reservation or order starts, and how many finish. The gap between started and completed is where the money is. If a hundred people start and a third finish, you have a booking flow problem and no amount of visibility will fix it.

These four move weeks before the dining room does. That is their entire value.

Which numbers look important and are not?

The ones that are easiest to report, which is not a coincidence.

Followers. A follower count records past interest, not present demand. It goes up when you post well and stays up when you close.

Impressions and reach. Impressions tell you a screen displayed something. It is the least useful number a platform gives you and usually the largest, which is why it appears first in every report.

Rank position on one keyword. Local rank varies by where the searcher stands, so one position from one point is a screenshot, not a measurement.

Website traffic in total. Total sessions blend people looking for your hours with people looking for a job. The page level numbers mean something. The total does not.

Anything you cannot act on by Friday. If a number goes up and you do not know what to do differently, it is scenery.

One more caution. Do not try to A/B test your way to answers at restaurant traffic levels. Conversions per variant is about 16, from 2 × (1.96 + 0.84)², over the square of the lift you want: 100 for 40%, 400 for 20%derived. Running it anyway produces confident conclusions from noise.

Should I trust a local pack click-through rate table?

No, and this is the single most confidently fabricated number in local marketing.

You will be shown a table. Position one gets 17.6% of clicks, position two 15.4%, position three 15.1%. Or 17.8, 13.6 and 10.4. Or 24.4, 13.3 and 8.6. Three different tables, all presented as research.

Here is where each comes from.

The 17.6 / 15.4 / 15.1 set is FirstPageSage, an SEO agency, which describes its own work as "a meta-analysis, combining research on click-through rates" with no sample size and no method, sourced to a list of other blogs. One of those blogs now cites FirstPageSage back. The 17.8 / 13.6 / 10.4 set is attributed to a "BrightLocal Local Pack CTR Study" that does not exist: the URL returns a 404 and no such report is in BrightLocal's research index. The 24.4 / 13.3 / 8.6 set is credited to a "Moz Local CTR Study" that also does not exist, published by the same aggregator as the fake BrightLocal one.

The structural reason nobody has this number is permanent. Google exposes no local pack position dimension anywhere. The Business Profile performance report publishes interactions, search terms, views, direction requests, calls, website clicks, messages, bookings, products, menus and offers. No rank, no position, no pairing of impressions to clicks by rank. Search Console has no local pack dimension either. The only way to produce that table is a lab test or a clickstream panel that can see the whole result page.

One real attempt exists. BrightLocal ran a Mechanical Turk study in October 2018 in which 5,500 testers looked at five static search results. With Local Services Ads present, local pack position one took 16.1% of clicks, position two 10.4%, position three 2.3%, and all three together 28.8%. Without the ads, 17.3, 12.0, 2.9 and 32.3%.

Every caveat is load-bearing. Paid microworkers clicking a screenshot, not observed behavior. Five queries, all San Francisco, all service-area categories, desktop only. October 2018, on a result page that has since absorbed local pack ads, Local Services Ads and AI local packs, which is to say the page that study measured no longer exists. Its own data undercuts a pure position model too: in one test position two outperformed position one, and the stated top reason for clicking was the review rating in the local pack at 15.44%, ahead of position. Its author wrote in a forum thread that "nothing is as accurate as Google's own search data... for now, this is the best I can do."

So never convert a ranking change into a traffic forecast. Measure the change in actions on your own profile across a 90 day window, pair it with grid rank data so the ranking movement is documented separately, and report the two as correlated rather than as a formula. Any proposal that hands you a CTR table has told you something important about the person who wrote it.

How far ahead do these numbers actually move?

Weeks, not days, and that is the point.

Review pace is the slowest and the most predictive. A profile that stops collecting reviews looks fine for a while, then starts sliding, and by the time you feel it in the dining room the gap took a month or two to open.

Direction requests and calls move faster, usually within a few weeks of a change to your listing or your hours. They are the first thing to fall when something breaks on the profile, which makes them a useful alarm.

Menu page visits move fastest of the four, because they respond to visibility changes almost immediately.

So the reading rhythm is monthly, on the same day, compared against your own previous months. Not against a benchmark somebody published for a different market with a different concept. Your own trend is the only honest comparison, and the same discipline applies to the catering numbers a restaurant owner should track if that side of the business matters to you.

How do I tell a real move from noise?

By using a rolling window instead of a single comparison. Here is the arithmetic.

Say your last four months of new reviews were 3, 5, 2 and 4, and the restaurant ranking above you posted 6, 5, 7 and 6.

Month to month that says almost nothing. A two-review gap in March is weather, a busy week, one manager on vacation.

Rolled up, it says everything.

• You: 14 reviews in four months, or 3.5 a month.
• Them: 24 in four months, or 6 a month.
• Gap: 2.5 a month, which is 30 a year.

That is a real signal, and it is invisible at the monthly level. Whitespark's benchmarking rule tells you what to do with it: match your competitor's rate and add one, which here means a target of seven a month, not three.

The same discipline kills the worst report line in the category. If a vendor tells you your "AI visibility score" went from 40% to 45% this week, that is not a result. Researchers at the University of St. Gallen measured how unstable these systems are: identical prompts run ten times on the same day shared only 32% to 43% of their cited sources, and two consecutive days shared only 34% to 42%. Their prescription is seven or more runs per prompt per day and a two to four week rolling window, never week over week, because at seven runs the standard error is about 0.10. A five point weekly move sits inside the noise.

They also found ChatGPT activated web search on only about 42.2% of runs, so most runs cited nothing at all.

Know what Google itself publishes, too. Its generative AI performance data in Search Console gives impressions, pages, countries, devices and dates. Clicks, click-through rate, position and the user's prompt are not available, and clicks from AI Overviews arrive in your analytics as ordinary organic traffic with no separate channel. Nobody can hand you an AI click number for your restaurant, because Google does not publish one.

How do I read these without hiring anyone?

One sheet, five lines, once a month, same day.

Write down: direction requests, calls, menu page visits, new reviews this month, and new reviews for the two restaurants ranking above you. That last one takes five minutes of counting and it is the most valuable line on the page.

Add booking or order starts and completions if your platform will tell you. If it will not, that is worth knowing about your platform, and it is one of the tradeoffs in what reservation platforms do to your visibility.

Keep four months visible so the rolling comparison above is possible without extra work.

Do not add more lines. A sheet with twenty numbers gets read once and then stops getting read, and the numbers that matter drown in the ones that do not.

Then connect it to the operation. Marketing numbers tell you what is arriving. They do not tell you whether you made money on it. A month with more covers and worse margin is a worse month, and the place that usually shows up is scheduling, which is the subject of finding the four points of labor variance you are losing.

What to do this week

Open your Business Profile performance report and write down direction requests and calls for the last three months. That is your baseline.

Find your menu page number in analytics. If you do not have analytics installed, install it. That is a one hour job and it is the difference between reading and guessing.

Count new reviews for the last four months, for yourself and for the two profiles above you in your main search. Twelve numbers total. Roll them up.

Make the five line sheet, put a date on it, and give it a permanent place next to whatever you already review monthly.

Then leave it alone for thirty days. These numbers are meaningless week to week and checking them daily will make you reactive, which is the opposite of the point. The underlying mechanics behind all of them are the same ones in how restaurants actually get found on Google.

Questions about your numbers?

Email me at eric@seod.com and tell me which numbers you currently look at every month, even if the answer is none. I will tell you which of them actually lead cover count, which are scenery, and the one line I would add to your sheet. One reply, no dashboard, no login required.

I ran restaurants for sixteen years and I answer these myself. If you are already tracking the right things, I will tell you to stop reading and go back to the floor.

Or keep reading more on restaurant marketing.