Open-source agentic AI Search Execution Platform. Track AI visibility, find opportunities, and let agents turn signals into prioritized actions and drafted fixes across ChatGPT, Claude, Gemini, Google AI Overviews, Google AI Mode, Perplexity, Grok, and Microsoft Copilot.
TypeScript aeoai-agentai-searchanswer-engine-optimizationbrand-monitoringchatgptclaudecopilotdockergeminigeogoogle-ai-overviewsgrokmcpnextjsopen-sourceperplexityseosupabasetypescript
-- ============================================================================
-- 00226: the Content Editor's reads, bounded by their window
-- ============================================================================
-- Four content reads, redefined after they were timed on production (the
-- reviews of #1287, #1292, #1308 and #1309). Each returns what it did; each
-- reads less to get there.
--
-- * content_page_overlap (00207): the window's answers come from
-- idx_prompt_results_brand_created_cover alone (an index-only scan), and
-- the prompts are picked where the citations meet them. A prompt filter
-- on the answers made the planner combine the prompt_id index with the
-- brand's and read about 16k heap blocks a call.
-- * content_draft_sources (00211) read every answer each source prompt
-- ever had, through the prompt_id index, and the citations' primary key
-- once per answer (8 s cold for a 92-prompt draft):
-- - the window's answers come from the cover index alone, for a draft
-- with prompts as well as one with a page, and the source prompts
-- are picked from them after;
-- - each prompt's answers are counted on that index; those naming the
-- brand come from the Mentions rollup (mentions_prompt_daily, 00157)
-- for the window's whole UTC days, and from the answer rows of its
-- first, partial day only;
-- - the fan-outs and the cited pages are read from each prompt's
-- latest 30 answers in the window, so "searched N×" and a page's
-- citations count those: they are the only answer rows fetched whole
-- (search_queries), and their citations a fifth of the window's.
-- * content_plan_pages (00220, as 00221 redefined it): the overlaps' pairs
-- take their window from the current answers alone. A date bound on the
-- citations too is estimated from prompt_result_citations' last analyze
-- (no index leads with created_at), at about one row for recent days,
-- and that plan rescanned the own addresses once per citation (15–19 s
-- at 7 days on the largest brand).
-- * content_plan_insights (00221) joined every address the brand had cited
-- in the window to citation_urls before keeping its own and the top
-- 1,000. Now the own ids come from citation_urls' domain index first, the
-- rows are picked on the rollup's url ids, and only those picked are read
-- from citation_urls. The own list keeps the 5,000 cited most (a brand on
-- a large platform's domain has more), and each call is planned for its
-- own values (plan_cache_mode), as content_plan_pages is.
--
-- The index-only scans stay off the heap once prompt_results and
-- prompt_result_citations are vacuumed daily (00224), and that migration's
-- daily analyze keeps the planner's estimate of a recent window current.
-- Grants and comments carry over from 00207–00221 (create or replace); the
-- insights' comment is restated for its cap.
create or replace function public.content_page_overlap(
p_brand_id uuid,
p_prompt_ids uuid[],
p_since timestamptz default now() - interval '30 days'
)
returns table (
page text,
title text,
prompts_in_common integer,
citations bigint,
gsc_position numeric
)
language plpgsql
stable
security definer
set search_path to 'public'
set plan_cache_mode to 'force_custom_plan'
as $$
begin
if not exists (
select 1 from public.brands b
join public.profiles pf on pf.organization_id = b.organization_id
where b.id = p_brand_id and pf.id = auth.uid()
) and auth.role() is distinct from 'service_role' then
return;
end if;
if coalesce(cardinality(p_prompt_ids), 0) = 0 then
return;
end if;
return query
with own_domains as (
select lower(regexp_replace(bd.domain, '^www\.', '')) as domain
from public.brand_domains bd
where bd.brand_id = p_brand_id
),
own_urls as materialized (
select cu.id, cu.url, cu.title, lower(public.normalize_page_path(cu.url)) as page_key
from public.citation_urls cu
where cu.domain in (select domain from own_domains)
),
-- The window's answers, from the (brand_id, created_at) cover index
-- alone. The prompts are picked in `cited`: a prompt filter here makes the
-- planner combine the prompt_id index with the brand's and read the heap.
answers as materialized (
select pr.id, pr.prompt_id
from public.prompt_results pr
where pr.brand_id = p_brand_id
and pr.created_at >= p_since
and pr.platform is distinct from 'chatgpt-shopping'
),
cited as (
select ou.page_key, ou.url, ou.title, a.prompt_id
from public.prompt_result_citations prc
join answers a on a.id = prc.prompt_result_id
join own_urls ou on ou.id = prc.url_id
where prc.brand_id = p_brand_id
and prc.created_at >= p_since
and a.prompt_id = any(p_prompt_ids)
),
pages as (
select
c.page_key,
-- The address cited most often stands for the page.
mode() within group (order by c.url) as page,
max(c.title) as title,
count(distinct c.prompt_id)::integer as prompts_in_common,
count(*)::bigint as citations
from cited c
group by c.page_key
),
google as (
select
g.page_key,
round((sum(g.position * g.impressions) / nullif(sum(g.impressions), 0))::numeric, 1)
as position
from public.gsc_page_stats g
where g.brand_id = p_brand_id
and g.date >= (now() at time zone 'utc')::date - 28
and g.page_key in (select p.page_key from pages p)
group by g.page_key
)
select p.page, p.title, p.prompts_in_common, p.citations, gl.position
from pages p
left join google gl on gl.page_key = p.page_key
order by p.prompts_in_common desc, p.citations desc, p.page
limit 10;
end;
$$;
create or replace function public.content_draft_sources(
p_draft_id uuid,
p_days integer default 30
)
returns jsonb
language plpgsql
stable
security definer
set search_path to 'public'
set plan_cache_mode to 'force_custom_plan'
as $$
declare
v_brand_id uuid;
v_prompt_ids uuid[];
v_page_key text;
v_host text;
v_since timestamptz;
-- The window's first whole UTC day.
v_first_day date;
v_result jsonb;
begin
select d.brand_id, d.prompt_ids, lower(public.normalize_page_path(d.source_url)),
-- Lower-cased first: a scheme may be written HTTPS://.
regexp_replace(substring(lower(d.source_url) from '^https?://([^/:?#]+)'), '^www\.', '')
into v_brand_id, v_prompt_ids, v_page_key, v_host
from public.content_drafts d
where d.id = p_draft_id;
if v_brand_id is null or (
not exists (
select 1 from public.brands b
join public.profiles pf on pf.organization_id = b.organization_id
where b.id = v_brand_id and pf.id = auth.uid()
) and auth.role() is distinct from 'service_role'
) then
return null;
end if;
if p_days is null or p_days < 1 or p_days > 90 then
raise exception 'content_draft_sources: 1 to 90 days are needed' using errcode = '22023';
end if;
v_since := now() - make_interval(days => p_days);
v_first_day := (v_since at time zone 'utc')::date + 1;
-- A page on a site that isn't the brand's is no page of the brand's.
if v_host is null or not exists (
select 1 from public.brand_domains bd
where bd.brand_id = v_brand_id
and lower(regexp_replace(bd.domain, '^www\.', '')) = v_host
) then
v_page_key := null;
end if;
with own_domains as (
select lower(regexp_replace(bd.domain, '^www\.', '')) as domain
from public.brand_domains bd
where bd.brand_id = v_brand_id
),
-- The page's addresses: its path on its host.
page_urls as materialized (
select cu.id
from public.citation_urls cu
where v_page_key is not null
and cu.domain = v_host
and lower(public.normalize_page_path(cu.url)) = v_page_key
),
-- The window's answers, from idx_prompt_results_brand_created_cover alone
-- (an index-only scan). The prompts are picked after this read: a prompt
-- filter here makes the planner combine the prompt_id index with the
-- brand's and read every answer each prompt ever had.
window_answers as materialized (
select pr.id, pr.prompt_id, pr.created_at
from public.prompt_results pr
where (v_page_key is not null or coalesce(cardinality(v_prompt_ids), 0) > 0)
and pr.brand_id = v_brand_id
and pr.created_at >= v_since
and pr.platform is distinct from 'chatgpt-shopping'
),
-- The prompts whose answers cited the page the draft optimizes.
page_prompts as (
select distinct wa.prompt_id
from public.prompt_result_citations prc
join window_answers wa on wa.id = prc.prompt_result_id
where prc.url_id in (select pu.id from page_urls pu)
and prc.brand_id = v_brand_id
and prc.created_at >= v_since
),
source_prompts as (
select p.id, p.text, p.id = any(v_prompt_ids) as in_draft,
p.id in (select prompt_id from page_prompts) as cites_page
from public.prompts p
join public.prompt_sets ps on ps.id = p.prompt_set_id
where ps.brand_id = v_brand_id
and (p.id = any(v_prompt_ids) or p.id in (select prompt_id from page_prompts))
),
-- The source prompts' answers in the window, newest first per prompt.
answers as materialized (
select wa.id, wa.prompt_id, wa.created_at,
row_number() over (partition by wa.prompt_id order by wa.created_at desc) as recency
from window_answers wa
where wa.prompt_id in (select sp.id from source_prompts sp)
),
-- The answers naming the brand: the window's whole UTC days from the
-- Mentions rollup (mentions_prompt_daily, 00157), and its first, partial
-- day from those answers' rows. A run still in progress joins the rollup
-- when it completes.
named as (
select m.prompt_id, m.mention_answers as answers
from public.mentions_prompt_daily m
where m.brand_id = v_brand_id
and m.day >= v_first_day
and m.prompt_id in (select sp.id from source_prompts sp)
union all
select a.prompt_id, 1
from answers a
join public.prompt_results pr on pr.id = a.id
where a.created_at < v_first_day::timestamp at time zone 'utc'
and pr.mention_count > 0
),
prompt_counts as (
select a.prompt_id, a.answers, coalesce(n.mentioned, 0) as mentioned
from (select prompt_id, count(*) as answers from answers group by prompt_id) a
left join (
select prompt_id, sum(answers) as mentioned from named group by prompt_id
) n on n.prompt_id = a.prompt_id
),
-- The only answer rows read: each prompt's latest 30.
sampled as (
select pr.id, pr.platform, pr.search_queries
from answers a
join public.prompt_results pr on pr.id = a.id
where a.recency <= 30
),
fanout_rows as (
select
a.id as answer_id,
lower(trim(regexp_replace(q.item->>'query', '\s+', ' ', 'g'))) as norm,
trim(regexp_replace(q.item->>'query', '\s+', ' ', 'g')) as query,
coalesce(nullif(q.item->>'source_platform', ''), a.platform) as platform
from sampled a
cross join lateral jsonb_array_elements(
case when jsonb_typeof(a.search_queries) = 'array' then a.search_queries else '[]'::jsonb end
) as q(item)
where coalesce(trim(q.item->>'query'), '') <> ''
),
fanouts as (
select min(query) as query, array_agg(distinct platform order by platform) as platforms,
count(distinct answer_id) as searched
from fanout_rows
group by norm
order by count(distinct answer_id) desc, min(query)
limit 30
),
-- The citations of the same sample as the fan-outs, each prompt's latest
-- 30 answers: every answer in the window is 100k+ citations on a large
-- brand, read one answer at a time. Grouped by address first; only the
-- top 20 addresses are read.
cited_ids as (
select prc.url_id, count(*) as citations, count(distinct prc.prompt_result_id) as answers
from answers a
join public.prompt_result_citations prc
on prc.prompt_result_id = a.id
and prc.brand_id = v_brand_id
and prc.created_at >= v_since
where a.recency <= 30
group by prc.url_id
order by count(*) desc, count(distinct prc.prompt_result_id) desc, prc.url_id
limit 20
),
cited as (
select cu.url, cu.title, cu.domain, ci.citations, ci.answers
from cited_ids ci
join public.citation_urls cu on cu.id = ci.url_id
),
queries as (
select g.query, sum(g.clicks)::bigint as clicks, sum(g.impressions)::bigint as impressions
from public.gsc_query_stats g
where g.brand_id = v_brand_id
and g.date >= (now() at time zone 'utc')::date - 28
group by g.query
order by sum(g.clicks) desc, sum(g.impressions) desc, g.query
limit 300
)
select jsonb_build_object(
'prompts', coalesce((
select jsonb_agg(jsonb_build_object(
'id', sp.id,
'text', sp.text,
'inDraft', sp.in_draft,
'citesPage', sp.cites_page,
'answers', coalesce(pc.answers, 0),
'mentioned', coalesce(pc.mentioned, 0)
) order by sp.in_draft desc, sp.text)
from source_prompts sp
left join prompt_counts pc on pc.prompt_id = sp.id
), '[]'::jsonb),
'fanouts', coalesce((
select jsonb_agg(jsonb_build_object(
'query', f.query, 'platforms', to_jsonb(f.platforms), 'searched', f.searched
) order by f.searched desc, f.query)
from fanouts f
), '[]'::jsonb),
'citedPages', coalesce((
select jsonb_agg(jsonb_build_object(
'url', c.url, 'title', c.title, 'domain', c.domain,
'citations', c.citations, 'answers', c.answers
) order by c.citations desc, c.answers desc, c.url)
from cited c
), '[]'::jsonb),
'gscQueries', coalesce((
select jsonb_agg(jsonb_build_object(
'query', q.query, 'clicks', q.clicks, 'impressions', q.impressions
) order by q.clicks desc, q.impressions desc, q.query)
from queries q
), '[]'::jsonb),
'brandDomains', coalesce((select jsonb_agg(domain order by domain) from own_domains), '[]'::jsonb),
'competitorDomains', coalesce((
select jsonb_agg(distinct lower(regexp_replace(c.domain, '^www\.', '')))
from public.competitors c
where c.brand_id = v_brand_id and coalesce(c.domain, '') <> ''
), '[]'::jsonb)
) into v_result;
return v_result;
end;
$$;
create or replace function public.content_plan_pages(
p_brand_id uuid,
p_days integer default 30
)
returns jsonb
language plpgsql
stable
security definer
set search_path to 'public'
-- A cached generic plan can't see the brand's size (00207).
set plan_cache_mode to 'force_custom_plan'
as $$
declare
v_last date := (now() at time zone 'utc')::date - 1;
v_cur_day date;
v_prev_day date;
v_cur timestamptz;
v_until timestamptz;
v_result jsonb;
begin
if not exists (
select 1 from public.brands b
join public.profiles pf on pf.organization_id = b.organization_id
where b.id = p_brand_id and pf.id = auth.uid()
) and auth.role() is distinct from 'service_role' then
return null;
end if;
if p_days not in (7, 30, 90) then
raise exception 'p_days must be 7, 30 or 90';
end if;
v_cur_day := v_last - p_days + 1;
v_prev_day := v_cur_day - p_days;
v_cur := v_cur_day::timestamp at time zone 'utc';
v_until := (v_last + 1)::timestamp at time zone 'utc';
with own_domains as (
select lower(regexp_replace(bd.domain, '^www\.', '')) as domain
from public.brand_domains bd
where bd.brand_id = p_brand_id
),
own_urls as materialized (
select cu.id, cu.url, cu.title, lower(public.normalize_page_path(cu.url)) as page_key
from public.citation_urls cu
where cu.domain in (select domain from own_domains)
),
url_counts as materialized (
select d.url_id, d.platform, d.day >= v_cur_day as is_cur, sum(d.citations)::bigint as n
from public.citation_url_daily d
where d.brand_id = p_brand_id
and d.day >= v_prev_day
and d.day <= v_last
and d.url_id in (select id from own_urls)
and d.platform is distinct from 'chatgpt-shopping'
group by d.url_id, d.platform, d.day >= v_cur_day
),
url_totals as (
select
u.page_key,
u.url,
u.title,
coalesce(sum(c.n) filter (where c.is_cur), 0) as cur,
coalesce(sum(c.n) filter (where not c.is_cur), 0) as prev
from url_counts c
join own_urls u on u.id = c.url_id
group by u.page_key, u.url, u.title
),
pages as materialized (
select
t.page_key,
-- The address cited most often stands for the page.
(array_agg(t.url order by t.cur + t.prev desc, t.url))[1] as page,
max(t.title) as title,
sum(t.cur)::bigint as cur,
sum(t.prev)::bigint as prev
from url_totals t
group by t.page_key
order by sum(t.cur + t.prev) desc, t.page_key
limit 100
),
answers as (
select
r.platform,
coalesce(sum(r.answers) filter (where r.day >= v_cur_day), 0)::bigint as cur,
coalesce(sum(r.answers) filter (where r.day < v_cur_day), 0)::bigint as prev
from public.citation_result_daily r
where r.brand_id = p_brand_id
and r.day >= v_prev_day
and r.day <= v_last
and r.platform is distinct from 'chatgpt-shopping'
group by r.platform
),
per_platform as (
select
u.page_key,
c.platform,
coalesce(sum(c.n) filter (where c.is_cur), 0) as cur,
coalesce(sum(c.n) filter (where not c.is_cur), 0) as prev
from url_counts c
join own_urls u on u.id = c.url_id
where u.page_key in (select p.page_key from pages p)
and c.platform is not null
group by u.page_key, c.platform
),
down as (
-- A platform tracked in both windows, citing the page at a lower rate.
select
pp.page_key,
array_agg(
pp.platform
order by pp.prev::numeric / a.prev - pp.cur::numeric / a.cur desc, pp.platform
) as platforms
from per_platform pp
join answers a on a.platform = pp.platform
where a.prev > 0
and a.cur > 0
and pp.prev::numeric / a.prev > pp.cur::numeric / a.cur
group by pp.page_key
),
cur_answers as materialized (
select pr.id, pr.prompt_id
from public.prompt_results pr
where pr.brand_id = p_brand_id
and pr.created_at >= v_cur
and pr.created_at < v_until
and pr.platform is distinct from 'chatgpt-shopping'
),
-- The window comes from cur_answers: a citation carries its answer's
-- created_at. A date bound on prc too would be estimated from
-- prompt_result_citations' last analyze (no index leads with created_at),
-- at about one row for recent days, and that plan rescans own_urls once
-- per citation.
page_prompts as (
select distinct u.page_key, a.prompt_id
from public.prompt_result_citations prc
join own_urls u on u.id = prc.url_id
join cur_answers a on a.id = prc.prompt_result_id
where prc.brand_id = p_brand_id
and u.page_key in (select p.page_key from pages p)
),
competing as (
select a.page_key, b.page_key as other
from page_prompts a
join page_prompts b on b.prompt_id = a.prompt_id and b.page_key <> a.page_key
group by a.page_key, b.page_key
having count(*) >= 2
),
page_overlaps as (
select cm.page_key, count(*) as pages
from competing cm
group by cm.page_key
),
snippet_days as (
select distinct (l.created_at at time zone 'utc')::date as day
from public.ai_traffic_logs l
where l.brand_id = p_brand_id
and l.visit_of is null
and l.created_at >= v_cur
and l.created_at < v_until
),
visits as (
select lower(public.normalize_page_path(l.url)) as page_key, 1::bigint as n
from public.ai_traffic_logs l
where l.brand_id = p_brand_id
and l.visit_of is null
and l.created_at >= v_cur
and l.created_at < v_until
union all
select lower(public.normalize_page_path(g.landing_page)), g.sessions::bigint
from public.ga_ai_traffic_stats g
where g.brand_id = p_brand_id
and g.sessions > 0
-- A session without a landing page would read as the home page.
and g.landing_page <> ''
and g.date >= v_cur_day
and g.date <= v_last
and not exists (select 1 from snippet_days s where s.day = g.date)
),
page_visits as (
select v.page_key, sum(v.n) as visits
from visits v
where v.page_key in (select p.page_key from pages p)
group by v.page_key
)
select jsonb_build_object(
'days', p_days,
-- The earlier window's first day, and the brand's first tracked day: a
-- window before it is only partly tracked, or not at all.
'prevFrom', v_prev_day,
'trackedSince', (
select min(r.day) from public.citation_result_daily r where r.brand_id = p_brand_id
),
'answers', (
select jsonb_build_object(
'cur', coalesce(sum(a.cur), 0),
'prev', coalesce(sum(a.prev), 0)
)
from answers a
),
'pages', coalesce((
select jsonb_agg(
jsonb_build_object(
'page', p.page,
'title', p.title,
'citations', jsonb_build_object('cur', p.cur, 'prev', p.prev),
'platformsDown', coalesce(to_jsonb(d.platforms[1:3]), '[]'::jsonb),
'aiVisits', coalesce(pv.visits, 0),
'overlaps', coalesce(o.pages, 0)
)
order by p.cur + p.prev desc, p.page
)
from pages p
left join down d on d.page_key = p.page_key
left join page_overlaps o on o.page_key = p.page_key
left join page_visits pv on pv.page_key = p.page_key
), '[]'::jsonb)
)
into v_result;
return v_result;
end;
$$;
create or replace function public.content_plan_insights(
p_brand_id uuid,
p_days integer default 30
)
returns jsonb
language plpgsql
stable
security definer
set search_path to 'public'
-- A cached generic plan can't see the brand's size (00207).
set plan_cache_mode to 'force_custom_plan'
as $$
declare
-- Whole days to yesterday, as the Plan's pages (00220).
v_last date := (now() at time zone 'utc')::date - 1;
v_day date;
v_from timestamptz;
v_until timestamptz;
v_result jsonb;
begin
if not exists (
select 1 from public.brands b
join public.profiles pf on pf.organization_id = b.organization_id
where b.id = p_brand_id and pf.id = auth.uid()
) and auth.role() is distinct from 'service_role' then
return null;
end if;
if p_days not in (7, 30, 90) then
raise exception 'p_days must be 7, 30 or 90';
end if;
v_day := v_last - p_days + 1;
v_from := v_day::timestamp at time zone 'utc';
v_until := (v_last + 1)::timestamp at time zone 'utc';
with own_domains as (
select lower(regexp_replace(bd.domain, '^www\.', '')) as domain
from public.brand_domains bd
where bd.brand_id = p_brand_id
),
-- The brand's own addresses, through citation_urls' domain index, as
-- content_plan_pages' own_urls (00220).
own_ids as materialized (
select cu.id
from public.citation_urls cu
where cu.domain in (select domain from own_domains)
),
cited as materialized (
select d.url_id, sum(d.citations)::bigint as citations
from public.citation_url_daily d
where d.brand_id = p_brand_id
and d.day >= v_day
and d.day <= v_last
and d.platform is distinct from 'chatgpt-shopping'
group by d.url_id
),
chosen as (
(
select c.url_id, c.citations, true as own
from cited c
where exists (select 1 from own_ids o where o.id = c.url_id)
order by c.citations desc, c.url_id
limit 5000
)
union all
(
select c.url_id, c.citations, false
from cited c
where not exists (select 1 from own_ids o where o.id = c.url_id)
order by c.citations desc, c.url_id
limit 1000
)
),
top_urls as (
select ch.url_id, ch.citations, ch.own, cu.url, cu.title
from chosen ch
join public.citation_urls cu on cu.id = ch.url_id
),
snippet_days as (
select distinct (l.created_at at time zone 'utc')::date as day
from public.ai_traffic_logs l
where l.brand_id = p_brand_id
and l.visit_of is null
and l.created_at >= v_from
and l.created_at < v_until
),
visits as (
select count(*)::bigint as n
from public.ai_traffic_logs l
where l.brand_id = p_brand_id
and l.visit_of is null
and l.created_at >= v_from
and l.created_at < v_until
union all
select coalesce(sum(g.sessions), 0)::bigint
from public.ga_ai_traffic_stats g
where g.brand_id = p_brand_id
and g.sessions > 0
and g.date >= v_day
and g.date <= v_last
and not exists (select 1 from snippet_days s where s.day = g.date)
),
analytics as (
select
coalesce(sum(g.key_events), 0)::bigint as key_events,
coalesce(sum(g.transactions), 0)::bigint as transactions,
coalesce(sum(g.purchase_revenue), 0)::numeric as revenue
from public.ga_ai_traffic_stats g
where g.brand_id = p_brand_id
and g.date >= v_day
and g.date <= v_last
)
select jsonb_build_object(
'days', p_days,
'totalCitations', (select coalesce(sum(c.citations), 0) from cited c),
'citations', coalesce((
select jsonb_agg(
jsonb_build_object(
'url', t.url,
'title', t.title,
'own', t.own,
'citations', t.citations
)
order by t.citations desc, t.url
)
from top_urls t
), '[]'::jsonb),
'results', (
select jsonb_build_object(
'visits', (select coalesce(sum(v.n), 0) from visits v),
'keyEvents', a.key_events,
'transactions', a.transactions,
'revenue', a.revenue,
'analytics', (
select nullif(b.ga_property_id, '') is not null
from public.brands b where b.id = p_brand_id
),
'purchases', a.transactions > 0 or a.revenue > 0 or exists (
select 1 from public.ga_channel_stats c
where c.brand_id = p_brand_id
and c.date >= v_day
and c.date <= v_last
and (c.transactions > 0 or c.purchase_revenue > 0)
)
)
from analytics a
)
)
into v_result;
return v_result;
end;
$$;
comment on function public.content_plan_insights(uuid, integer) is
'Content Editor (CE-014): the addresses AI answers cite for the brand over the last N (7, 30 or 90) whole UTC days to yesterday (from citation_url_daily): its own pages (the 5,000 cited most) and the 1,000 others cited most, with the total; and what AI-referred visitors did on the site: visits, Google Analytics AI key events, purchases and revenue. Security definer with an organization check.';