What Should Your SEO Agency Deliver Each Month?

Judge an SEO retainer on three lines kept apart: changes you can open and check yourself, signals that moved, and booked jobs by location. Impressions show visibility, including AI Overview and AI Mode appearances. Read them alongside calls, enquiries and jobs so the report explains both who is seeing the business and what that attention produces.

Last verified: November 1, 2026.

On sales calls this fall, owners kept asking us some version of "What are the monthly deliverables?" Google's guide to hiring an SEO, last updated June 5, 2026, puts it more sharply: "Will you share with me all the changes you make to my site...?" Ask it every month, before you pay the invoice.

These checks work whoever does your marketing (staff, an agency or a done-for-you partner like Cheers), because they run on pages and accounts you already own. Weight booked jobs most.

This month's report covers October, so start this week with one email. Ask whoever runs your marketing for a dated list of every change made that month, each with a link you can open; any vendor keeping records can send that within a day or two. A page URL counts, and "content optimization" doesn't.

While you wait, open your Business Profile on Google Maps and write down the date of the newest owner photo (Photos, then By owner) and the newest Update (the Updates tab). If either predates the retainer, ask what the photo and Update plan is.

Three lines, kept apart

Ask for this as the first page of every monthly report, or keep it in a sheet yourself, with three columns for each item: this month, last month and the same month last year. Leave a cell blank when you don't have the number, because a missing number isn't zero.

What shipped. Items on the dated change list, items you opened and confirmed, and items that didn't check out (and why).

Signals. Business Profile calls (call-button taps) and direction requests, by location; Search Console clicks to service and location pages; AI answers, named in __ of __ tracked questions, by engine and city; and AI Overview and AI Mode impressions (Generative AI report; already inside the Web total).

Booked jobs. By location, and by lead source ("how did you hear about us").

Line one asks whether work happened, line two whether Google, Maps and the AI engines noticed it, and line three whether the phone rang with jobs you could book. Don't fold them into one score, because a blended number hides which part is working. When they disagree, check the measurement before the work, then let booked jobs decide.

Say impressions, map rankings and AI mentions climb all quarter while clicks fall and booked jobs hold flat. Before anyone blames the work, find out whether the call tracking or the number on the profile changed partway through, since either change breaks the before-and-after comparison. If both held steady, go back to the change list and ask what each item was supposed to move.

Each line moves on its own schedule, and how long AI search work takes to show up shows which one should have moved by now.

Line one: what shipped, checked on the live site

You can open every item below yourself, and most take a minute or two. Where we have a base rate from the wider market, it's at the end of the item, so you can tell an unusual gap from a common one.

New or rewritten pages. Open each URL on the change list, then enter it in Search Console's URL Inspection bar to see whether Google has crawled it. A dated link to a live page about a service or location you sell counts; "content optimization" with no URL doesn't, while crawl and index status belong on the separate search-uptake line. A live page is delivered before Google necessarily crawls it.

Blog or news posts. Check your blog's index page for the newest date and the images on the last few posts. Count the delivered post separately from its quality. Check whether the topic, dates and images fit the agreed brief; generic images or a missing visible date are quality issues, not proof that no work shipped.

Service and location pages tied to profiles. On each Business Profile, tap Website. Each profile should open its own location's page, and every profile pointing at the homepage doesn't count.

LocalBusiness schema. Run the homepage and one location page through validator.schema.org. You're looking for LocalBusiness markup with that location's name, address, phone and hours, matching the page; "schema added" with no page named doesn't count. Base rate: 54.6% of contractor homepages carry it, and 47.0% carry it with the phone number inside.

Text that loads without scripts. In Chrome, View page source (Ctrl+U, or Cmd+Option+U on a Mac) and search for your phone number and the page's first sentence, or use URL Inspection, then View crawled page. Services, city and phone should be in the source; if key content depends on scripts, check the rendered result with URL Inspection. JavaScript alone does not make a delivered page invalid. Base rate: 5.8% of contractor homepages showed under 150 words before JavaScript ran.

The phone number. Tap it on your homepage on a phone and compare it with each profile, then ask ChatGPT a question your customers ask and read the number it prints. You want one clickable number per location that matches the profile, or a written reason it differs, rather than a tracking number nobody told you about. Base rate: 86.0% of contractor homepages have a clickable phone link, and 17.6% of the US numbers ChatGPT printed weren't on the business's own site in the pages we checked.

Newest owner photo. In Google Maps, open your profile, then Photos, then By owner, and note the newest date. Dated photos of real jobs, crews or trucks from this month count; stock images, or a photo count with no dates, don't.

Newest Update. Note the date on your profile's Updates tab. A dated Update about a real service, offer or season counts, and "posts published" with no links doesn't. Google's merchant rules allow offers and your own contact details, but not unrelated contact details or links.

Services, categories, hours and description. Ask for dated before-and-after screenshots, then open Edit profile and confirm what's live. The changes should match what you sell and when you're open, with none of the limited-time promotions or "call now" lines Google's merchant rules keep out of the description; "profile optimized" on its own doesn't count.

Review replies. Sign in to Google Maps, open your profile's Reviews, sort by Lowest and check every one- to three-star review from the month. Each one needs a reply that mentions the job; a reply count with no links, or the same template on every review, doesn't count.

Directory listings. Open each listing the report names and compare name, address, phone and hours with the profile. A link to each corrected listing counts, and "citations built" with no links doesn't.

AI answers. The same questions, cities and engines each month, each with the date and the answer text or a screenshot, recorded as named or not per question, engine, city and date. "AI visibility improved" with no question, engine or date doesn't count, and neither does your company name typed into ChatGPT.

Base rates come from Cheers' Nov. 1, 2026 read of 976 readable US contractor homepages from Google's local map results, and the ChatGPT figure from the Cheers Market Baseline, our own research panel. Method has the limits.

An llms.txt file, a plain-text summary some sites publish for AI tools, is a one-time job and doesn't belong on a monthly list. If your site serves one, check that it names the right business and phone number.

Line two: signals, read for what they count

Google's Business Profile help page defines calls as "the number of times a customer clicked on the call button on your Business Profile." That's a tap, not a connected call or a booked job, so keep those taps apart from your phone and booking records. Which Business Profile metrics matter by location covers the rest of those definitions.

Google is also testing a redesigned performance report, but Search Engine Roundtable's Oct. 23 write-up mentions nothing about AI answers or calls placed by AI, so don't expect it to fill that gap.

In Search Console, read clicks to your service and location pages before total impressions, since part of an impressions jump can be AI answer appearances nobody clicked. The Generative AI performance report shows how many, by page, country, date and device. Those impressions already sit inside the Web total, so don't add them to it. Google says it rolled the report out to all websites as of Aug. 31, 2026, though it may not appear for a site with too few AI feature impressions.

For AI answers, record whether you were named for each tracked question, with the engine, city and date. If you don't track any questions yet, Cheers' free AI visibility check gives you a starting list. Leave tone scores off the KPI list, because tone almost never turns negative in answers about who to hire: our automated labels scored 29 of 70,624 business mentions in US Cheers Market Baseline answers in October 2026 as negative. Skip screenshots of ChatGPT describing your company, too; typing your name in almost always gets a decent description back, which says nothing about whether you make the shortlist.

Read October's AI referral numbers with care, because the tags came apart that month. Sistrix found ChatGPT tagged every source link with a utm_source tag set to chatgpt.com up to Oct. 8, 2026, then only 1% to 24% a day from Oct. 9 to 15, and links in the new format pass no referrer. Search Engine Roundtable reported on Oct. 30 that Gemini stopped adding a utm_source tag set to gemini around Sept. 28 and, citing a practitioner's tests, that the Gemini iOS app sends no referrer. So some of October's ChatGPT and Gemini visits sit under Direct in your channel report, and the AI referral count can come in low; how local service brands should track AI search traffic has the analytics setup.

What's inside your impressions: a two-minute check

Run this on your own Search Console property before the next report meeting; if you can't log in, that's the first thing to ask for. Google's filter help page documents the steps.

Open Performance, then Search results, and set the date range to the month the report covers. Click "+ Add filter", choose Queries, then "Custom (regex)". Enter `^\S+(\s+\S+){7,}$`, which matches searches of eight or more words, click Apply and note the impressions total. Then change the regex to `.` and note impressions again; a single dot matches every query Google shows you, so both totals leave out the same anonymized searches. Divide the first number by the second.

We ran the same rule on 19 websites from 11 US home-trade companies that work with Cheers, mostly plumbing and HVAC. On the typical site, searches of eight or more words made up 2.0% of impressions from Oct. 1 to 14, 2026; Method has the spread across sites and the pooled share.

Our own site looks nothing like that. On cheers.tech, which writes about AI search, the rule caught 29.6% of impressions in the same two weeks, and those long searches almost never produced a click. One of them, "what options exist for tracking ai search visibility across chatgpt, gemini, and perplexity simultaneously?" drew about 1,900 impressions and no clicks between July and October.

Who runs searches like that? Search Console doesn't record it. Many read like prompts written for an AI tool rather than words typed into Google, and Google counts AI Mode searches in Search Console, so some likely come from people and software, AI-tracking tools among them, asking AI Mode questions. That's an inference, not something the data shows.

If your share lands in the low single digits, like most of the sites we measured, long searches don't explain an impressions jump, so check the Generative AI report and clicks to your service and location pages instead. If it runs far higher, the way cheers.tech's does, open the long queries and ask what they are before anyone counts them as progress.

Line three: booked jobs, by location and lead source

Booked jobs don't come from any marketing tool, which is why they belong on page one. Pull them from your CRM or field service software by location and month, with the same month last year beside them.

Lead source got harder to read in October. In the Cheers Market Baseline, 12.6% of US ChatGPT local answers carried at least one phone number from Oct. 2 to 8, 2026, and 92.5% did from Oct. 9 to 31. A customer who calls directly from that answer can reach the branch without creating a website session, which makes call and intake records part of AI attribution.

Have whoever answers the phone ask "how did you hear about us" on every booked call and log the answer with the job. That gives you something to hold up against the source your reporting software assigns.

Call tracking adds a wrinkle: where the business's site ran a call-tracking script, 41.0% of the numbers ChatGPT printed for it in October 2026 weren't on the business's own website in the pages we checked, against 12.1% when the site didn't run one. Those calls may land on a number your report credits to another source, such as the Business Profile or a directory, or to none; the study can't tell which.

If your site runs call tracking, put a few of your customers' questions to ChatGPT, then ask whoever runs your reporting which source gets credit for the number it prints. Should home services use call tracking for AI search covers where tracking numbers belong.

What to ask for before the next report

Put these four in writing before the next report is due; the October change-list email already covers the first. Ask for the dated change list for the month, with a link to every change, and for your own login to Search Console and to every Business Profile, if you can't get in today. Then ask for the exact questions, cities and engines they track in AI answers, with the raw answers, and for booked jobs by location if they have CRM access, or a note that you'll supply them.

Google's hiring guide has the follow-up, and you want the answer in writing: "What kind of results do you expect to see, and in what timeframe? How do you measure your success?" Judge every report against that timeframe rather than a ranking, since the same page warns that "No one can guarantee a #1 ranking on Google."

For an audit, the guide says to "only grant read access to Search Console." Either way, your company should own the Business Profile, the analytics account, the domain and the website code.

If you run many locations, ask for the same four items location by location. And if the answers don't come back before your renewal date, what to use instead of a local SEO agency compares the delivery models; this page only checks the work.

The 1,096-site agent-readiness census gives the website checks below a dated market baseline.

For October reporting, account for the ChatGPT referral-tracking interruption before treating a traffic decline as a fall in demand.

The 1,117-site website and Maps study helps separate observed website features from unsupported claims that a design choice causes better rankings.

Before accepting a visibility claim, use the AI report reliability guide to check its questions, dates, denominators and repeated observations.

Method

Search Console data covers 19 websites from 11 US home-trade companies that work with Cheers, mostly plumbing and HVAC, plus cheers.tech, for Oct. 1 to 14 and Oct. 15 to 27, 2026. We counted only impressions Search Console ties to a query. A search counts as long at eight or more words separated by spaces. We report customers only in aggregate and never name them. On cheers.tech, long searches have run at about half of non-branded impressions from June through October.

Across the 19 sites from Oct. 1 to 14, the middle half sat between 0.5% and 3.7%, 3 of the 19 reached 5% or more, and the pooled share across all 19 was 1.8%. From Oct. 15 to 27 the share was about 2% again, and long searches drew too few clicks on these sites to compare with short ones.

One company accounts for more than half of the impressions in the 19-site set, so we report the typical (median) site first and the pooled share second. These are established Cheers customers, mostly multi-location, in Cheers' markets, so a one-truck shop's site may look different. Two October windows don't make a trend.

We left out searches that exactly match a question Cheers tracks in AI answers; leaving them in moves none of the shares here by more than a point.

We added up rows page by page, while Search Console's Queries tab counts each search once for the whole site (Google explains the difference), so your number can differ from ours by up to about a point.

Website base rates: on Nov. 1, 2026, Cheers read the homepages of 1,096 US home-service contractor websites drawn from Google's local map results, with Cheers customers removed; 976 homepages were readable as served. We read the homepage only, as raw HTML with no JavaScript run, so presence rates, including the schema and phone-link rates, are floors, and anything on inner pages or loaded by script can slip past the check. We drew the sites from Cheers' weekly scans of Google's local map results in markets where 10 Cheers home-service customers operate, so they aren't a national random sample.

AI answer numbers come from the Cheers Market Baseline. The Cheers Market Baseline is our own research panel: every week it asks ChatGPT, Google AI Mode and Gemini the same 32 home-service questions in 100 metro areas, 60 in the US and 40 in Canada. This page uses the 60 US metros, October 2026. Tone labels are automated, so treat the negative count as approximate. The Oct. 9 to 31 phone-number share counts 5,835 of 6,307 US ChatGPT local answers.

The phone comparisons use ChatGPT answers from Oct. 2 to 31, 2026, checked against each business's website on Nov. 1, and only numbers we could check: 66 of 374 numbers for the 17.6% figure, 32 of 78 where the site ran a call-tracking script and 34 of 280 where it didn't. A number missing from the pages we checked tells you only that we didn't find it there, and none of these figures counts calls or booked jobs.

Sources

• AI Features and Your Website, Google Search Central, updated Dec. 10, 2025. AI Overviews and AI Mode count inside Search Console's Web totals.
• Do You Need an SEO? Tips for Hiring an SEO, Google Search Central, updated June 5, 2026. The hiring-guide questions and warnings quoted here.
• Schema Markup Validator, Schema.org, undated. The tool for checking a page's LocalBusiness markup.
• Maps user-contributed content policy: Advertising & solicitation, For merchants, Google, undated. What Updates and profile descriptions may carry.
• Understand your Business Profile performance & insights, Google Business Profile Help, undated. Google's definition of a call.
• Google Business Profiles Tests New Advanced Performance Reports, Search Engine Roundtable, Oct. 23, 2026. The redesigned performance report test.
• Generative AI performance report (Search), Search Console Help, undated. The report's rollout, coverage and breakdowns.
• What changed in ChatGPT in October: GPT-6, UTM tracking and sources, Sistrix, Oct. 30, 2026. ChatGPT's October tag and referrer change.
• Google Gemini Stopped Attaching UTM Parameters To Cited URLs, Search Engine Roundtable, Oct. 30, 2026. Gemini's tag stop and the iOS app's missing referrer.
• Performance report (Search results): Advanced filtering and comparison, Search Console Help, undated. The regex filter steps.
• Performance report (Search results): Overview and basic setup, Search Console Help, undated. Page versus property totals.

Sources read November 1, 2026.

Dylan Allen-Arnegård is the CEO & Co-Founder of Cheers, the done-for-you local visibility partner for established home-service companies and PE-backed portfolios.