← Rating: vet clinics in Dallas
Ratings · Methodology

How we measure digital visibility

Version 3.1 of the methodology behind IGM’s digital presence ratings. It covers markets where maps, search and AI answers come from Google and OpenAI. The first release under it covers veterinary clinics in Dallas.

Version 3.1Fixed October 1, 2026Index out of 90

This edition of the methodology is for markets where maps, search and AI models are Google and OpenAI services. Releases for Russia follow version 2.1. Versions are never mixed, and each release page names its own.

The methodology is fixed with a date before data collection. It does not change while a release is in progress. If a rule turns out to be a poor one, the release still follows it. The fix goes into the next version.

Version history. Version 3.0 was fixed on September 25, 2026, before data collection for the first release. Version 3.1 was fixed on October 1, 2026, after that collection. It sets the review threshold by a formula, keeps AI systems outside the index as “experimental” columns and adds rules for the release page. The first release (veterinary clinics in Dallas, data collected September 28–29, 2026) was recalculated under version 3.1 without any new measurements, and its page says so.

1. What the rating measures, and what it doesn’t

The rating measures a company’s digital presence. That means how visible the company is to someone looking for a provider in maps, in search and in AI answers. It also covers the state of the company’s website as a way to bring in clients.

Data comes from public sources and is collected from the outside, without any involvement from the companies being rated.

The rating does not assess the quality of the core service. We don’t judge the quality of medical care, teaching, construction or legal help. A high index doesn’t mean a company treats patients better. A low index doesn’t mean it treats them worse. It means one thing only: the company was easier or harder to find on the measurement date.

The rating is entertainment and educational content based on our own method. It is not an expert assessment and not a reputation rating. Rows are ordered solely by the calculated index; no ordinal places are awarded. Rows less than one point apart are shown as level: the gap is within measurement error.

2. Industry and geography

One rating covers one industry in one city. The industry is determined by the primary category of the Google Maps listing and by the company’s website.

The city boundary is the postal city in the listing’s address. The sample includes organizations whose listing address shows the rating’s city. For the first release, that is Dallas, TX. Neighboring cities in the metro area are not included, even when they share a border with it. They have their own postal city. The rule is mechanical and can be checked against the address.

One caveat: in the US, a postal city sometimes extends past the city’s legal boundary. Such addresses stay in the sample. A customer sees the same city on the listing that we do.

3. How the sample is built

Selection is mechanical. The rule is published before data collection and does not change during a release.

Candidates come from Google Maps. We take the candidates returned by Google Maps under our grid-and-saturation protocol. Saturation shows that the protocol stopped finding new places, not that every business in the market was found. One Google Maps search shows no more than 120 places. These are not only places in the visible area, but also the nearest ones across the city. So we cover the city with a grid of map windows by coordinates and merge the lists by listing ID. Completeness is checked by saturation: the last windows of the grid should add no new places. If they do, the grid is made denser until it saturates.

Why the sample is not built from search results. Selecting by the same trait that is measured later takes away that measure’s power to tell companies apart. It also hides the key finding: what share of the industry is not visible at all.

Industry. The primary category of the listing. For veterinary clinics, these are Veterinarian, Animal hospital, Veterinary care and Emergency veterinarian service. Grooming, pet stores, shelters, boarding, dog daycare and laboratories are not included.

Inclusion criteria. The listing is active, not marked Permanently closed or Temporarily closed. The company’s own website on its own domain returns code 200 on the measurement date. The number of Google Maps reviews is at or above the threshold set for the release.

Exclusion criteria.

  • No website of its own. This includes a page on a directory, on social media or on a site-builder subdomain instead (*.wixsite.com, *.square.site, *.godaddysites.com and similar).
  • The website is unavailable on the measurement date.
  • The listing is closed.
  • The business does not match the industry.
  • A solo practice, under the rule below.
  • A different postal city.

Solo practices. The rating compares companies, not individual professionals. The test is mechanical. The listing name is a person’s first and last name, with or without a title (Dr., DVM, DDS). It has no word for a business. Business words include hospital, clinic, center, care, practice, group, services and the like.

  • “Jane Smith DVM” and “Dr. John Brown” are solo practices and are not included.
  • “Dr. Smith’s Animal Hospital”, “Smith Veterinary Clinic” and “Brown & Lee Animal Care” are companies. The name comes with a business word, so it is a brand.
  • Some listings belong to individual doctors at a clinic, such as “Oak Street Vet: Jane Smith DVM”. After the colon comes a person’s name with a title. These are listings of professionals, not companies, and they are not included. The clinic takes part with its own listing.
  • A surname alone is a brand.
  • A person’s name in the domain does not matter.
  • Names with a title are filtered out mechanically. Two-word names with no title and no business word go to a person for review, together with the selection funnel.

Chains. One brand, one row. Branches are grouped by website domain. We measure the listing with the most reviews, the flagship. Branch reviews are not added together. Branches have different ratings, different review recency and a different share of owner replies.

Review threshold. The threshold comes from a formula, not from a person’s choice: the highest of 500, 300, 200, 150, 100, 75 or 50 Google reviews that still leaves at least 40 companies in the final sample. It is published together with the number of organizations dropped at each step. If even 50 reviews leaves fewer than 40 companies, the threshold is 50 and the sample is published as is.

Clients of the organizer are marked in the sample. Their data is calculated on the same terms as everyone else’s.

4. What we measure and how

All measurements come from a single address, within one week. They use the website’s home page unless stated otherwise. Services run in English (hl=en), with the country set to the US (gl=us). The location is the rating’s city, set with the uule parameter and the canonical name “Dallas,Texas,United States”.

We measure from outside the US. A pilot on September 25, 2026 checked what that changes. With uule set, Google returned Dallas city results, with “Dallas, TX” in the location line.

4.1. Maps presence — 25 points

IndicatorHow it is measuredPoints
ListingsA Google Maps listing and a company page on Yelp. The Yelp page is found through Yelp search by name and city. It counts if the phone or the address matches.both: 4; one: 2; none: 0
CompletenessCategory, business hours, phone, website link, at least five photos in the listing gallery. Also at least five attributes on the About tab: accessibility, services, amenities, parking and so on.all: 4; one or two missing: 2; more missing: 0
Number of reviewsThe review count on the Google Maps listing. The scale is relative. For each company we take the share of participants in the release that have strictly fewer reviews.up to 20%: 1; 20–40%: 2; 40–60%: 3; 60–80%: 4; 80% and up: 5
Average ratingThe listing’s average rating4.7 and up: 4; 4.3–4.7: 3; 4.0–4.3: 2; lower: 1
Review recencyDate of the latest review, sorted by “Newest”up to 1 month: 3; 1–3 months: 2; 3–12 months: 1; older: 0
Handling reviewsShare of reviews with a “Response from the owner” among the latest twentyover half: 3; a quarter to a half: 2; under a quarter: 1; no replies: 0
Contact matchThe phone or the address from the listing appears on the websitematch: 2; mismatch: 0

Section maximum: 25 points.

What counts as a review. Google Maps shows a single review count, and it includes ratings without text. The rating takes that number as it is. Recency and the reply share are counted on the same set: the “Newest” feed. There, ratings without text appear alongside written reviews, and the owner can reply to any of them. So the count, the recency and the replies in this section all refer to the same thing.

Why the review scale is relative. In version 2.1 the scale was fixed: over 300 reviews earned 5 points. The entry threshold for the sample turned out to be higher than the threshold for full points. Every participant got 5 out of 5, and the indicator stopped telling companies apart. A relative scale always tells them apart. Equal values get equal points, and the result does not depend on the order of rows. That is why review points are not comparable across releases. Ratings for different cities and industries are not compared anyway.

Why attributes rather than a description. Google Maps does not show a business description on the listing. We checked three Dallas clinics on September 25, 2026 in a signed-in profile. The “From the owner” block on the overview turned out to be a dated post. “From the business” on the About tab turned out to be a group of attributes. As with Yandex Maps in September 2026, no points are taken off for a field the platform does not show. Attributes take its place, like “Features” in version 2.1, with the same threshold of five.

Relative dates. Google shows a review’s age in words. We convert it to days as follows:

  • “N minutes ago”, “N hours ago”, “a day ago” and “N days ago”: the number of days stated.
  • “a week ago”: 7 days; “N weeks ago”: 7×N.
  • “a month ago”: 30 days; “N months ago”: 30×N.
  • “a year ago”: 365 days; “N years ago”: 365×N.

The “up to 1 month” band runs to 31 days inclusive, so “a month ago” falls within it.

Search results are captured once for each of the twenty core queries. The same twenty lists are used to check every participant, by matching domains.

The core is 20 commercial queries for the industry. It is based on Google Keyword Planner data for the rating’s city. Keyword Planner reports the volume of a phrase together with its close variants. For example, “vet dallas”, “veterinarian dallas” and “dallas vet” form one group with one volume. So the core takes one phrase per group. A group is a set of phrases with the same volume and the same monthly curve over the year. We take the shortest wording. Otherwise the same demand would enter the core twice.

Branded queries and adjacent niches are excluded. Queries with “near me” and without a city name are included, since they carry most of the demand.

The core takes the twenty groups with the highest volume. When volumes tie at the edge of the top twenty, the wording is sorted alphabetically. The rule is mechanical. In Dallas, five phrases with a volume of 90 competed for one spot. The choice between them should not depend on who builds the core.

Query weight is the average monthly Keyword Planner volume over the year. Keyword Planner volume and Yandex Wordstat frequency are different measures and are not compared.

What counts as a result. The first ten organic Google results. Not counted: ads (Sponsored), the map pack, AI Overview, “People also ask”, videos, images and “Things to know”. The result’s domain is taken from the visible address line under the title.

Location. All queries are run from one point: the rating’s city, set with uule. We tested this in a pilot on September 25, 2026. “vet near me” searched from central, north and south Dallas returned the same nine organic domains. The point inside the city only affects the map pack, which is not part of this section.

Score: share of demand covered. A spot in the top three takes the query’s full volume. A spot in the top ten takes half, and anything lower takes zero. The total is divided by the total volume of the core and multiplied by 20. The table shows the number of top-3 and top-10 hits and the share of demand covered.

4.3. Presence in AI answers — 15 points

Systems: ChatGPT (chatgpt.com), Google AI Overviews (the block above Google results) and Google AI Mode (the AI answer mode in Google Search). The release page lists the systems used.

Questions: 10 questions in English, in everyday language, that a person asks when choosing a provider in the industry. Each question is asked 10 times in each system. AI Overviews and AI Mode run with the rating’s city as the location (uule). ChatGPT runs in a regular session, with the city name in the question.

A mention is the company’s name in the answer text, including source captions and place cards inside the answer. The company’s own domain in the answer text also counts. Names are normalized: case, quotation marks, legal forms (LLC, Inc, PLLC, PC, PA), the DVM title, and “&” versus “and” do not matter. A chain counts by its brand. A mention of a directory such as Yelp does not count for the company.

“No AI block” and “company not named” are different things. Google does not show an AI Overview for every query. A run with no AI block is void. It does not enter the denominator, just like a failed ChatGPT run. A run with an AI block that does not name the company is valid. It counts as zero for that company.

When a system drops out. Sometimes the AI block appears in fewer than 30 of 100 runs. Then the system is left out of this section for every company. The release page says: “block shown in X of 100 runs, not counted.” This rule applies to the system as a whole and does not remove the section. A share based on twenty or thirty runs still tells companies apart. A share based on a handful of runs is noise. A system outside the index stays on the release page as an “experimental” column with its shares. A release may also leave a system out by decision, for example when the AI block appeared on only some of the questions. The release page then gives the reason.

Score. The share of mentions across the combined pool. We add up the runs with a mention across the systems in the calculation. Then we divide by the sum of their valid runs. Each system’s result is published in its own column and is not averaged across systems.

Share of mentionsPoints
over 30%15
10–30%10
1–10%5
0%0

AI answers are not deterministic. A single answer proves nothing, so we publish only the share across many runs.

4.4. Website health — 20 points

This section is unchanged from version 2.1.

IndicatorHow it is measuredPoints
Server responseTime to first byte for the home page, averaged over three measurements at least 60 seconds apartup to 600 ms: 4; 600–1200: 3; 1200–2500: 1; over: 0
Page weightSize of the home page response in kilobytes, without loaded resourcesup to 200 KB: 3; 200–500: 2; 500–1000: 1; over: 0
Content without JavaScriptWord count in the HTML before scripts run. We remove the script, style, noscript and svg tags, then all markup. The rest is split by spaces.over 300 words: 4; 100–300: 2; under 100: 0
Mobile versionA viewport meta tag, and no horizontal scrolling at a window width of 375 pixelsboth checks: 3; one: 1; none: 0
Address consolidationFour pairs: with and without www, http and https, with and without a trailing slash, with and without the .html extension. We count the pairs where both versions return 200 instead of a redirect.0 duplicates: 3; 1: 2; 2: 1; more: 0
Canonical linkA link with rel canonical on the home page and on five random inner pages. Pages come from the sitemap, or from links on the home page if there is no sitemap. The source is recorded in the measurement data.everywhere: 2; partly: 1; none: 0
Service filesrobots.txt is available, does not block everything and points to the sitemap. The sitemap is available and contains valid XML.both fine: 1; otherwise: 0

Section maximum: 20 points.

Sites without inner pages. Some sites have no sitemap and no internal links: a one-page site, or a shell drawn entirely by a script. For such a site, the canonical link is checked on the only page that exists. “Everywhere” then means “on the home page.” The case is noted in the measurement data.

Response time from outside the US. We measure server response from outside the US. A pilot on September 25, 2026 showed that Dallas websites respond through CDN nodes in Europe. The distance does not move them across thresholds. The spread between measurements of the same site is larger.

4.5. Machine-readability — 10 points

IndicatorHow it is measuredPoints
Organization markupThe Organization type or a relevant subtype in JSON-LDpresent: 3; absent: 0
Markup depthNumber of unique type values across all JSON-LD blocks on the home page5 or more: 3; 2–4: 2; 1: 1; none: 0
BreadcrumbsThe BreadcrumbList typepresent: 1; absent: 0
File for AI modelsllms.txt returns code 200present: 1; absent: 0
Address and phone in markupThe organization block has address and telephone filled inboth: 1; otherwise: 0
Open GraphOpen Graph meta tags on the home pagepresent: 1; absent: 0

Section maximum: 10 points. The weight of llms.txt went down from two points to one. It is our own criterion. For a local business, the address and phone in the markup matter more.

5. The final index

SectionMaximum
Maps presence25
Search visibility20
Website health20
Presence in AI answers15
Machine-readability10
Total90

A section that could not be measured for even one company is dropped from the calculation for all participants. A failed measurement is not a zero. Zero means “we checked, and it isn’t there.” An unavailable website is checked again: http only, a parked domain, DDoS protection, a 429 response. Otherwise the participant leaves the sample and is counted among those not included. It never gets a zero.

6. Highlights and wording

Highlights and wording follow the same rules as version 2.1. Everything said about a named company is a measured fact, with the parameter and the date. For example: “for our twenty queries, the site did not appear in the top ten.” We don’t use evaluative words about named companies. Critical remarks are only about the market as a whole, with no names. Companies left out of the table are given as a number, without names or reasons.

A release may point out the highest value in a section. Such a note always names the section and the measurement date. For example: “the highest share of mentions in AI answers among the clinics in the sample, as of the measurement date.” Superlatives appear only together with the name of the measured parameter, in the same phrase.

We don’t use assessments of the quality of a company’s core service. We don’t draw conclusions about its good faith or guess at the reasons behind its numbers. We don’t compare companies on criteria outside this methodology.

7. Disputes, corrections and removal

Any company in the table can dispute its data. We reply within 5 business days and repeat the measurement the same way. If an error is confirmed, we publish the correction within 30 days, with the date of the change. A company is removed from the table at its first request, within 2 business days, no questions asked. Every change is listed in the release’s changelog.

Write to general@igm-marketing.com.

8. Disclosure

The organizer is IGM Professional, a marketing agency. We sell search, maps and AI-visibility services to local businesses, including businesses like the ones we rate. That is a conflict of interest, so the rules are fixed: rows, their order and corrections are not for sale; clients get no special treatment in the data; a participant that is our client on the publication date is marked as such. The organizer does not work in the industries it rates.

The IGM team collected the data with tools we built ourselves. They open Google Maps, Google Search, Yelp and ChatGPT in a regular browser and send requests to the websites’ public addresses from our European measurement node. Search volumes come from Google Keyword Planner.

9. Parameters of the first release

  • City: Dallas, TX. Search location: uule “Dallas,Texas,United States”, hl=en&gl=us.
  • Industry: veterinary clinics, with the categories listed in section 3.
  • Search volume: Google Keyword Planner, location “Dallas (US, City)”, language English.
  • Measurement window: one week. The release page gives the measurement dates for each section.

Questions about the method or about your company’s data? Write to general@igm-marketing.com. The first release is here: veterinary clinics in Dallas.