Datasets:

sempite
/

ai-visibility-index

# -*- coding: utf-8 -*-
"""AI Visibility Index: did an AI answer name this Google Maps business?
Matcher v2 (2026-09-17). Replaces the v1 rule inside ai_visibility_index.py,
which counted a business as "mentioned" when generic words from its listing
title appeared in the answer: the city ("Miami"), the trade ("Real Estate",
"Personal Injury", "Plumber") or "NYC". Those words are in nearly every answer
to "I need a [category] in [city]", so v1 overstated visibility.
v2 rules, in order:
  1. Split the Maps title into segments on " - ", " | ", ":", "," and "(".
     Titles are often keyword-stuffed ("Royal Lane Dental Center - Best Dentist
     in Dallas | Clear Aligners"), and the business name is the first segment
     that carries a distinctive word.
  2. Distinctive words = words that are not trade, city, legal-suffix or filler
     words (GENERIC below), at least 2 characters long.
  3. Distinctive words that are everyday English (COMMON: "one", "modern",
     "central") do not count on their own. Of the rest, one must appear as a
     whole word if the name has one, and at least two if it has two or more.
  4. If every distinctive word is COMMON, a two-word phrase from the name must
     appear ("one real", "modern dentistry", "four seasons").
  5. If no segment has a distinctive word ("Philadelphia Dental"), the whole
     first segment must appear as an exact phrase, minus a trailing "Inc"/"LLC".
  6. The business's own website domain appearing in the answer also counts.
Text is lowercased, apostrophes dropped ("Nick's" -> "nicks") and every other
non-alphanumeric character treated as a space, on both sides. Street addresses
are removed from the answer first, so "Bellaire Air Conditioning" does not match
a different company listed at "5933 Bellaire Blvd".
Ambiguous near-names ("Dental Arts of Dallas" for "Dallas Dental Arts") are left
counted as mentions: every judgement call errs toward visibility, so published
invisibility rates can only be understated by the matcher, not inflated.
The module is deliberately dependency-free so it can be published alongside the
data and re-run by anyone.
"""
import re

MATCHER_VERSION = "v2-2026-09-17"

GENERIC = set("""
plumber plumbers plumbing drain drains drainage sewer sewers rooter rooters water heater heaters
leak leaks pipe pipes repair repairs
dentist dentists dental dentistry orthodontics orthodontist orthodontists implant implants
cosmetic family pediatric general smile smiles teeth tooth dds dmd dr md
lawyer lawyers attorney attorneys law legal injury injuries personal accident accidents car truck
auto firm llp esq trial
real estate realty realtor realtors agent agents agency broker brokers brokerage homes home
properties property
hvac heating cooling air conditioning ac furnace furnaces heat mechanical
contractor contractors
inc llc pllc pc co company corp corporation group team associates partners services service
solutions pros pro professional professionals experts expert specialists specialist
center centre clinic office offices care
the and of in at by for a an to your my our near me best top local affordable quality
emergency hour hours 24 7 247 same day fast
son sons brothers bros
new york nyc ny manhattan brooklyn queens bronx
los angeles la socal dtla ca california
chicago il illinois loop
houston tx texas
phoenix az arizona
philadelphia philly pa pennsylvania
san antonio diego
dallas
miami fl florida brickell
downtown midtown uptown city usa us com
st on with by brokered
""".split())

# Everyday English words that appear in business names but also in ordinary
# prose ("one of the best", "a modern office", "a central location"). Alone they
# prove nothing, so a name whose only distinctive words are in this list must
# appear as a two-word phrase from the name ("one real", "modern dentistry").
COMMON = set("""
one two three four first all same main central public direct green blue red white black gold
golden silver modern american national united west east north south northwest southwest
northeast southeast found night tech master rescue mission elite temp happy relax ultra king
royal core champion village pride express star sky metro flow preferred heritage collective
insider weather makers seasons coastal mesa tree heights oak broad walnut street avenue arts
fifth premier choice smart bright prime premium advanced integrity trusted reliable honest
precision perfect superior summit peak liberty freedom eagle lion diamond crystal pacific
atlantic desert valley river lake park bay beach coast harbor sun sunset sunrise
lane road way point square plaza hills hill grove place
""".split())

_SEG_SPLIT = re.compile(r"\s+[-\u2013\u2014|]\s+|\s*[|:,(]\s*")


def norm(s):
    s = (s or "").lower().replace("'", "").replace("’", "")
    s = re.sub(r"[^a-z0-9]+", " ", s)
    return " " + " ".join(s.split()) + " "


def _words(seg):
    return norm(seg).split()


def name_core(name):
    """(segment, distinctive words) for the segment used to match."""
    segs = [s for s in _SEG_SPLIT.split(name or "") if s and s.strip()]
    for s in segs:
        d = [w for w in _words(s) if w not in GENERIC and len(w) >= 2]
        if d:
            return s.strip(), d
    return (segs[0].strip() if segs else (name or "")), []


def _root_domain(domain):
    if not domain:
        return None
    d = domain.lower().split("//")[-1].split("/")[0]
    if d.startswith("www."):
        d = d[4:]
    return d or None


# Street addresses ("5933 Bellaire Blvd", "1601 Walnut St") appear in search-backed
# answers next to OTHER businesses; a name that shares a street name must not match them.
_ADDRESS = re.compile(
    r" \d+[a-z]? (?:[nsew] )?(?:[a-z0-9]+ ){1,4}?(?:st|street|ave|avenue|blvd|boulevard|rd|road|dr|drive|"
    r"ln|lane|pkwy|parkway|hwy|highway|way|ct|court|pl|place|ter|terrace|cir|circle|fwy|freeway|sq|square)"
    r"(?= )")


def mentioned(business, answer, domain=None):
    """True/False, or None when there is no answer (engine excluded, never a miss)."""
    if not answer:
        return None
    text = _ADDRESS.sub(" ", norm(answer))
    dom = _root_domain(domain)
    if dom and dom in (answer or "").lower():
        return True
    seg, d = name_core(business)
    strong = list(dict.fromkeys(w for w in d if w not in COMMON))
    if strong:
        hits = sum(1 for w in strong if f" {w} " in text)
        return hits >= min(2, len(strong))
    phrase = _common_phrase(seg, d) if d else " ".join(_trim(_words(seg)))
    return bool(phrase) and f" {phrase} " in text


_TRAIL = {"inc", "llc", "pllc", "pc", "co", "corp", "corporation", "company", "llp"}


def _trim(words):
    while words and words[-1] in _TRAIL:
        words = words[:-1]
    return words


def _common_phrase(seg, d):
    """Shortest phrase from the name that is specific enough: the run of
    distinctive words if it is 2+ words long, otherwise the distinctive word
    plus its neighbour in the name."""
    w = _trim(_words(seg))
    i = next(k for k, x in enumerate(w) if x in d)
    j = i
    while j + 1 < len(w) and w[j + 1] in d:
        j += 1
    if j > i:
        return " ".join(w[i:j + 1])
    if i + 1 < len(w):
        return " ".join(w[i:i + 2])
    return " ".join(w[max(0, i - 1):i + 1])


def explain(business, domain=None):
    seg, d = name_core(business)
    strong = list(dict.fromkeys(w for w in d if w not in COMMON))
    if strong:
        rule = "need %d of %s" % (min(2, len(strong)), sorted(strong))
    elif d:
        rule = "phrase: %r" % _common_phrase(seg, d)
    else:
        rule = "phrase: %r" % " ".join(_trim(_words(seg)))
    return {"segment": seg, "distinctive": d, "domain": _root_domain(domain), "rule": rule}

	# -- coding: utf-8 --
	"""AI Visibility Index: did an AI answer name this Google Maps business?

	Matcher v2 (2026-09-17). Replaces the v1 rule inside ai_visibility_index.py,
	which counted a business as "mentioned" when generic words from its listing
	title appeared in the answer: the city ("Miami"), the trade ("Real Estate",
	"Personal Injury", "Plumber") or "NYC". Those words are in nearly every answer
	to "I need a [category] in [city]", so v1 overstated visibility.

	v2 rules, in order:
	1. Split the Maps title into segments on " - ", " \| ", ":", "," and "(".
	Titles are often keyword-stuffed ("Royal Lane Dental Center - Best Dentist
	in Dallas \| Clear Aligners"), and the business name is the first segment
	that carries a distinctive word.
	2. Distinctive words = words that are not trade, city, legal-suffix or filler
	words (GENERIC below), at least 2 characters long.
	3. Distinctive words that are everyday English (COMMON: "one", "modern",
	"central") do not count on their own. Of the rest, one must appear as a
	whole word if the name has one, and at least two if it has two or more.
	4. If every distinctive word is COMMON, a two-word phrase from the name must
	appear ("one real", "modern dentistry", "four seasons").
	5. If no segment has a distinctive word ("Philadelphia Dental"), the whole
	first segment must appear as an exact phrase, minus a trailing "Inc"/"LLC".
	6. The business's own website domain appearing in the answer also counts.

	Text is lowercased, apostrophes dropped ("Nick's" -> "nicks") and every other
	non-alphanumeric character treated as a space, on both sides. Street addresses
	are removed from the answer first, so "Bellaire Air Conditioning" does not match
	a different company listed at "5933 Bellaire Blvd".

	Ambiguous near-names ("Dental Arts of Dallas" for "Dallas Dental Arts") are left
	counted as mentions: every judgement call errs toward visibility, so published
	invisibility rates can only be understated by the matcher, not inflated.

	The module is deliberately dependency-free so it can be published alongside the
	data and re-run by anyone.
	"""
	import re

	MATCHER_VERSION = "v2-2026-09-17"

	GENERIC = set("""
	plumber plumbers plumbing drain drains drainage sewer sewers rooter rooters water heater heaters
	leak leaks pipe pipes repair repairs
	dentist dentists dental dentistry orthodontics orthodontist orthodontists implant implants
	cosmetic family pediatric general smile smiles teeth tooth dds dmd dr md
	lawyer lawyers attorney attorneys law legal injury injuries personal accident accidents car truck
	auto firm llp esq trial
	real estate realty realtor realtors agent agents agency broker brokers brokerage homes home
	properties property
	hvac heating cooling air conditioning ac furnace furnaces heat mechanical
	contractor contractors
	inc llc pllc pc co company corp corporation group team associates partners services service
	solutions pros pro professional professionals experts expert specialists specialist
	center centre clinic office offices care
	the and of in at by for a an to your my our near me best top local affordable quality
	emergency hour hours 24 7 247 same day fast
	son sons brothers bros
	new york nyc ny manhattan brooklyn queens bronx
	los angeles la socal dtla ca california
	chicago il illinois loop
	houston tx texas
	phoenix az arizona
	philadelphia philly pa pennsylvania
	san antonio diego
	dallas
	miami fl florida brickell
	downtown midtown uptown city usa us com
	st on with by brokered
	""".split())

	# Everyday English words that appear in business names but also in ordinary
	# prose ("one of the best", "a modern office", "a central location"). Alone they
	# prove nothing, so a name whose only distinctive words are in this list must
	# appear as a two-word phrase from the name ("one real", "modern dentistry").
	COMMON = set("""
	one two three four first all same main central public direct green blue red white black gold
	golden silver modern american national united west east north south northwest southwest
	northeast southeast found night tech master rescue mission elite temp happy relax ultra king
	royal core champion village pride express star sky metro flow preferred heritage collective
	insider weather makers seasons coastal mesa tree heights oak broad walnut street avenue arts
	fifth premier choice smart bright prime premium advanced integrity trusted reliable honest
	precision perfect superior summit peak liberty freedom eagle lion diamond crystal pacific
	atlantic desert valley river lake park bay beach coast harbor sun sunset sunrise
	lane road way point square plaza hills hill grove place
	""".split())

	_SEG_SPLIT = re.compile(r"\s+[-\u2013\u2014\|]\s+\|\s[\|:,(]\s")


	def norm(s):
	s = (s or "").lower().replace("'", "").replace("’", "")
	s = re.sub(r"[^a-z0-9]+", " ", s)
	return " " + " ".join(s.split()) + " "


	def _words(seg):
	return norm(seg).split()


	def name_core(name):
	"""(segment, distinctive words) for the segment used to match."""
	segs = [s for s in _SEG_SPLIT.split(name or "") if s and s.strip()]
	for s in segs:
	d = [w for w in _words(s) if w not in GENERIC and len(w) >= 2]
	if d:
	return s.strip(), d
	return (segs[0].strip() if segs else (name or "")), []


	def _root_domain(domain):
	if not domain:
	return None
	d = domain.lower().split("//")[-1].split("/")[0]
	if d.startswith("www."):
	d = d[4:]
	return d or None


	# Street addresses ("5933 Bellaire Blvd", "1601 Walnut St") appear in search-backed
	# answers next to OTHER businesses; a name that shares a street name must not match them.
	_ADDRESS = re.compile(
	r" \d+[a-z]? (?:[nsew] )?(?:[a-z0-9]+ ){1,4}?(?:st\|street\|ave\|avenue\|blvd\|boulevard\|rd\|road\|dr\|drive\|"
	r"ln\|lane\|pkwy\|parkway\|hwy\|highway\|way\|ct\|court\|pl\|place\|ter\|terrace\|cir\|circle\|fwy\|freeway\|sq\|square)"
	r"(?= )")


	def mentioned(business, answer, domain=None):
	"""True/False, or None when there is no answer (engine excluded, never a miss)."""
	if not answer:
	return None
	text = _ADDRESS.sub(" ", norm(answer))
	dom = _root_domain(domain)
	if dom and dom in (answer or "").lower():
	return True
	seg, d = name_core(business)
	strong = list(dict.fromkeys(w for w in d if w not in COMMON))
	if strong:
	hits = sum(1 for w in strong if f" {w} " in text)
	return hits >= min(2, len(strong))
	phrase = _common_phrase(seg, d) if d else " ".join(_trim(_words(seg)))
	return bool(phrase) and f" {phrase} " in text


	_TRAIL = {"inc", "llc", "pllc", "pc", "co", "corp", "corporation", "company", "llp"}


	def _trim(words):
	while words and words[-1] in _TRAIL:
	words = words[:-1]
	return words


	def _common_phrase(seg, d):
	"""Shortest phrase from the name that is specific enough: the run of
	distinctive words if it is 2+ words long, otherwise the distinctive word
	plus its neighbour in the name."""
	w = _trim(_words(seg))
	i = next(k for k, x in enumerate(w) if x in d)
	j = i
	while j + 1 < len(w) and w[j + 1] in d:
	j += 1
	if j > i:
	return " ".join(w[i:j + 1])
	if i + 1 < len(w):
	return " ".join(w[i:i + 2])
	return " ".join(w[max(0, i - 1):i + 1])


	def explain(business, domain=None):
	seg, d = name_core(business)
	strong = list(dict.fromkeys(w for w in d if w not in COMMON))
	if strong:
	rule = "need %d of %s" % (min(2, len(strong)), sorted(strong))
	elif d:
	rule = "phrase: %r" % _common_phrase(seg, d)
	else:
	rule = "phrase: %r" % " ".join(_trim(_words(seg)))
	return {"segment": seg, "distinctive": d, "domain": _root_domain(domain), "rule": rule}