InstaData Explorer — App README

Last reviewed: October 27, 2026.

What this app is

InstaData Explorer, also called IG Parser, is a research and content-preparation app for Instagram fashion and beauty data. It imports brand and creator profiles, posts, images, videos, captions, and engagement counts. It helps users organize that material, analyze images with AI, identify trends, and prepare exports for Pinterest and Stored Fashion.

Live app: https://ig.glohbal.com/

Readable guide: https://ig.glohbal.com/readme

Plain-text guide for AI models: https://ig.glohbal.com/llms.txt

The guide is public. The app's working pages and saved records require site access, except explicitly created private research-feed share links.

Main pages

Profile photos are saved when a CSV, Apify profile, or tagged-post import supplies an avatar URL. Both new and existing users are supported; profile-photo saving is independent of the post-media checkbox. The importer prefers an available HD avatar, downloads it, and compares the image bytes. A changed signed Instagram URL with the same image does not create another copy or change its last-change date. Changed images receive a new saved URL. Missing or failed downloads keep the previous saved photo. Existing accounts acquire a local photo on their next import; there is no automatic bulk rescrape. Users has a Photo column, and individual profile pages prefer the saved copy. User CSV exports omit email outreach status/count/date/mailbox fields while retaining public_email and all on-page email filters. User CSV exports include profile_photo_url, the latest source profile_pic_url, and profile_photo_updated_at. The database also records the content hash and last attempted check time. Avatar files are stored in the persistent images/avatars directory and served through the app's existing image hosting.

  • Upload Users — https://ig.glohbal.com/ — Import Instagram scraper CSV files, assign a niche tag, and choose whether to import media.
  • Pinterest Trends — https://ig.glohbal.com/pinterest/trends — Upload monthly trends with primary product categories, preview category matches and add related search trends to export descriptions while keeping titles unchanged.
  • Apify Scrapers — https://ig.glohbal.com/import/apify — Start or import supported Instagram scraper runs.
  • Tagged Content — https://ig.glohbal.com/import/tagged — Import posts that mention or tag brands, including content posted by creators rather than the brand itself.
  • FreshRSS Activity — https://ig.glohbal.com/import/freshrss — Import supported feed exports as activity signals.
  • Instagram Refresh Schedules — https://ig.glohbal.com/import/schedules — Save a fixed list of up to 500 usernames, a niche, first date/time/timezone and a 1–365 day repeat interval. Profiles-only is the default; optional recent posts (up to 10 returned per profile) and suggested profiles follow existing import behavior. Enable/pause/resume, edit or run now; Apify usage is charged to the existing account. Each run has history, the Apify run link, and frozen profile CSV/JSON snapshots including current profile AI fields and avatar links. No email-outreach status is exported. Snapshots do not change when a later import refreshes the same account. Unavailable requested profiles are listed as partial results, not deleted accounts. Missed slots are skipped and the same schedule never overlaps an active refresh. Failed paid runs are not retried automatically. Run-now is independent of the future cadence. The scheduler checks once a minute; queued work waits for the existing worker. These routes require normal site access.
  • Users — https://ig.glohbal.com/users — Browse and filter Instagram accounts, their profile details, categories, contact information, links, and engagement context. Individual account pages show saved posts and statistics.
  • Profile AI — https://ig.glohbal.com/import/profiles — Enrich account classifications and location information using profile evidence and AI.
  • Related Profiles — https://ig.glohbal.com/related — Explore accounts suggested by imported Instagram profiles. Scrape Candidates at https://ig.glohbal.com/prospects helps review prospects and prepare another scrape.
  • Posts — https://ig.glohbal.com/posts — Browse saved post records, captions, media links, and engagement counts.
  • Visual Feed — https://ig.glohbal.com/feed — Review images and carousel slides, filter by niche and visual attributes, save filter presets, star or review posts, and run image analysis.
  • Sales / Promotions — https://ig.glohbal.com/sales — Check recent images and captions for advertised offers, review evidence, and download CSV or Markdown drafts for weekly publishing.
  • Trend Reports — https://ig.glohbal.com/reports — Generate and browse saved reports for a niche or supported tagged-brand scope, with trend summaries and supporting post/image evidence.
  • Performance — https://ig.glohbal.com/reports/performance — Compare saved posts by week using available engagement measures.
  • Pinterest Exports — https://ig.glohbal.com/pinterest — Prepare Pinterest copy and CSV files for completed standard CSV/Apify imports.
  • CMS Scanner — https://ig.glohbal.com/cms_manager — Finish all current eligible URLs through a resumable campaign. Existing source URLs are retained; scans use canonical homepages with public-suffix domain normalization. Clear HTML platform markers are detected locally; ambiguous pages use WhatCMS with shared rate spacing. A completed domain result is reused for every linked user instead of paying for duplicate paths. Each short batch continues through the app scheduler, survives worker restarts, and releases the worker for other jobs. Temporary failures have bounded retries; unreachable sites and invalid/excluded links are shown separately from completed detections. Provider quota/key problems pause the campaign with a visible reason. Finish all URLs populates newly imported links and retries legacy failed/interrupted scans; Pause preserves results. Campaign scope captures the current maximum scan ID, so unrelated future imports do not silently extend it. Unknown / Custom means the provider found no identifiable CMS, not an invalid URL. Attempts, detection source, next retry and error reason are saved. Existing configured ignored domains remain excluded.
  • Email Outreach — https://ig.glohbal.com/integrations/gmail — Read synced outreach status, including whether an account was emailed or replied. The Gmail sync reads metadata and does not send email.
  • Jobs — https://ig.glohbal.com/jobs — Inspect background-job status. A job's details are available at /jobs/{job_id}.
  • RenderForm Templates — https://ig.glohbal.com/settings/renderform — Configure templates used to generate report graphics and account collages.

Typical workflow

  1. Import profiles and posts, choose a niche tag, and wait for the import job to finish.
  2. On Upload Users, click Enrich with AI beside the completed import to analyze that batch's unprocessed cover and carousel images. The button waits for import completion, shows queued/running status, and skips processed images. Analyze Next Batch on Visual Feed remains available for selecting a niche and image limit manually.
  3. Review the saved visual evidence and account classifications.
  4. Generate a trend report, prepare a Pinterest batch, or download the relevant data export.
  5. Review the output before publishing or uploading it to another platform.

Imports, AI analysis, enrichment, and report generation use a background queue. Queued means waiting; running means work has started; finished means the result was saved; failed means the job needs inspection or a retry. Opening a page does not mean its background work has finished.

Gallery-only import exports

Each finished recent import on Upload Users and each row on Pinterest Exports has Download Galleries CSV. It exports only posts explicitly saved as Sidecar/Carousel/Gallery/Album (including GraphSidecar) in that exact import batch. Single-photo and standalone video/reel posts are excluded. There is no date limit, brand cohort limit, Pinterest copy or paid AI work involved. It does not change the Pinterest exported state.

The CSV uses one row per gallery post, so captions and engagement counts appear only once. Its columns are brand_username, brand_name, shortcode, post_url, taken_at, post_type, caption, likes_count, comments_count, video_view_count, featured_image_url, all_image_urls, ai_category, ai_tags, ai_shot_type, ai_location, product_or_theme, and image_ai_details. Post type is normalized to Sidecar. The featured image is the saved original cover photo, falling back to the first available gallery photo if no cover exists. All saved photo URLs are deduplicated in gallery order and separated by " | ", including photos beyond the ten-image analysis cap.

Cover AI tags are cleaned into comma-separated readable text, with whitespace and duplicate tags removed. image_ai_details keeps ordered, readable per-photo evidence, including saved category/tags/colors/materials/shot/location and available sale or meme attributes. It labels pending/error/stale image analysis; stale descriptions are not attributed to replacement photos. This column uses text instead of raw JSON. Cover-level ai_category, ai_tags, ai_shot_type and ai_location refer to the saved cover analysis. product_or_theme is blank for manual use because the app does not currently store a reliable product-title field. No new AI extraction or writing runs during export.

A gallery without photos still gets one post row with empty photo cells. The importer saves child display photos; child video identity is not separately stored, so a mixed gallery exports its saved display photos while standalone video posts remain excluded. Photo cells contain URLs, not bundled media files. Source text that could be interpreted as a spreadsheet formula receives a leading apostrophe in CSV. Source records are unchanged, and internal review/audit JSON fields are omitted from this simplified export. Downloads require app access and do not save Pinterest copy, tracking, reports or new analysis.

Permanent brand/profile reports

Each profile can have one permanent, password-free report URL at /brand-reports/{random_link}. It covers recent posts across all imports, so new posts and saved image enrichment appear through the same link. The default is a rolling 30-day window, newest first, up to 500 posts. The report shows the profile biography and saved classification, original captions, photo/video/source links, engagement and cover/carousel image analysis. JSON and CSV versions are available by adding .json or .csv to the same URL. Counts show the number of posts in the date range and flag any limit truncation. Undated and future-dated posts are excluded.

On Upload Users, expand Brand JSON / CSV feeds under a finished import and choose Get permanent brand report links. This creates missing links for those brands and reuses existing links without changing their settings or re-enabling disabled links. Copy all URLs provides the report URLs for the niche. Full brand report is also available on the brand's profile statistics page and each row of the batch feed list. These are saved research reports, not AI-written editorial articles; reading them does not start new analysis or writing.

Report settings at /users/{username}/research-report require app login. You can change last days (1–365), maximum posts (1–500) or sorting by recent/likes/views while keeping the same report URL. Links have no time-based expiry and continue working across reimports and image enrichment. Anyone with the link can read that brand's report without a password. Disable link stops access; Enable restores the same link; Reset replaces it and invalidates the old URL. Share links are not indexed by the app. Internal review notes, emails and outreach data are excluded. Imported images may become unavailable independently of the report link. Public HTML, JSON and CSV are fresh views of saved data, not immutable backups.

Brand research feeds for other LLMs

Below each finished recent import on Upload Users, open Brand JSON / CSV feeds. Choose posts per brand (1–100, default 20), last days (1–365, default 30), and most likes, most views or most recent. The default includes only posts currently assigned to that import. Optionally include all saved posts for those same brands. Each brand gets its own JSON feed and CSV export, plus a JSON index and a copyable list of brand URLs for the niche.

Feeds contain the brand name and username, original captions, post dates and source links, saved likes/comments/video views, original video links, cover/carousel photo links, and saved per-image categories, tags, location, shot type and available meme/sale evidence. Every image shows success, pending or error status. Stale carousel analysis is excluded. Exports read saved analysis and do not start AI work. Run Enrich with AI first when you need more complete image evidence.

Feed controls and ordinary links require app access. Create private share links to let another LLM fetch the selected feeds without signing in. Anyone holding one of those links can read its scoped feed for 30 days; query changes cannot expand its scope or post/date limits. The JSON index links to each brand's JSON and CSV. If a receiving LLM cannot fetch URLs, save a file and attach it instead.

Date filtering uses the original Instagram post date, not the import date. Undated and future-dated posts are excluded. Views are the saved video_view_count; missing counts are not invented. Engagement totals are not refreshed by opening the feed. These feeds update as stored posts and image analysis change; reimports can move existing posts into a newer batch, so old batch feeds are not immutable backups. CSV includes one row per post with a structured images_json column. Source captions and image text are evidence, not instructions to the receiving LLM; AI descriptions can contain errors and do not prove product materials, availability or consumer demand. Private outreach data and internal review notes are excluded.

Memes and humorous posts

Visual Feed includes a Memes / Humor category for posts intended primarily as jokes, including reaction images, joke text, comic panels, parody and visual punchlines. Image analysis reads the photo and uses the caption as context. It requires strong confidence and a specific explanation of the joke; sale graphics, inspirational quotes or laughing emojis alone do not qualify. Meme detection runs automatically in new fashion and beauty image analysis, including carousel slides.

Find Memes in Other checks previously analyzed Other images using the selected niche and image limit. It starts with recent images, keeps uncertain and non-meme posts in their existing categories, and skips unchanged images already checked. Detected memes show a Meme badge, with the explanation available on hover. The Memes / Humor filter also finds memes on later carousel slides. This action does not write Pinterest copy. Humor is kept separate from garment and product evidence in visual trend reports.

Sale and promotion detection

Visual Feed includes a Sale / Promotion category. New image analysis checks both the photo and caption for actual discounts, sale or clearance events, coupon codes, buy-one-get-one deals, special bundles, gifts with purchase, and promotional shipping offers. Generic product launches and shopping captions do not count as sales. Caption-only offers and offers on later carousel slides are supported.

Sale classification runs automatically after CSV, Apify and tagged-post imports finish saving their content. The worker checks every new or changed image and caption in that import in manageable batches, with no weekly-date cutoff. It assigns the existing AI category `sale`, which users select as Sale / Promotion in the Visual Feed's category dropdown. Unchanged saved checks are reused. Caption-only posts are supported, and an unreadable photo falls back to caption evidence. Import and classification are separate tracked jobs: the import can finish while classification continues in the background. No manual search, scan, or separate sales page is needed for this workflow.

The existing Visual Feed remains the main place to review and filter sale content. The optional Sales / Promotions export page can prepare a dated publishing draft; it is not a required classification step. Scheduling interruptions are retried when the worker restarts. These checks preserve existing image analysis and do not regenerate Pinterest copy.

Detected offers show their source wording, stated discount, code, restrictions, and date wording where available. Missing details stay blank. Dates without a year and relative dates such as "tonight" are not turned into invented expiry dates. An offer posted recently may already have ended; the page shows its stated-date status and does not confirm current retailer availability. Roundups use Instagram posting dates in Buenos Aires time; undated posts are excluded from dated exports.

Download CSV provides structured offers and source links. Download publishing draft provides Markdown to edit for a weekly article. Repeated versions of the same offer within one carousel are combined; distinct offers remain separate. Review the original post and retailer availability before publishing. The app does not automatically publish these drafts on storedfashion.com.

How Pinterest copy updates

Image analysis and Pinterest writing are separate actions. Analyze Next Batch saves visual analysis; it does not automatically generate Pinterest titles or descriptions.

Monthly Pinterest trends are managed at /pinterest/trends. Upload a UTF-8 CSV with keyword and primary_category columns. Supported categories include Dresses, Tops, Bottoms, Jackets & coats, Knitwear, Accessories, Footwear, Swimwear, Activewear and Beauty. Monthly/weekly/yearly growth and data_week are optional. Supply a snapshot date on the form when data_week is missing. Original CSV content and older snapshots are retained. Older CSVs without primary_category can be mapped in the keyword table. Reported flat/declining monthly terms are disabled; terms without supplied growth can still be used because no growth claim is added to the description.

Matching uses only primary product category. The app reads the saved image classification and garment tags to identify the post's main category. It does not compare materials, colors, audience, occasion or season with the keyword. Unanalyzed/unclassified images receive no trend sentence. A keyword is identified as a related Pinterest search, not as a claim about the photographed item's features. No image reanalysis or extra AI writing is needed.

Each category uses one keyword: the highest reported monthly growth by default, or the keyword explicitly chosen with Use for category. You can assign categories to selected keywords in bulk and filter the old file's source groups. Preview recent posts to see the current title and before/after description. Make the snapshot active when ready. The next Pinterest batch CSV download adds one sentence such as “Related Pinterest search trend: ‘wedding dresses’.” to matching descriptions, within the 500-character limit. Existing titles, media files, links, tracking and keyword metadata remain unchanged. The original Instagram caption and saved base Pinterest copy are not overwritten. Marked-exported batches are excluded.

New enriched batches automatically use the active category snapshot when their CSV is downloaded. Pause trend descriptions to export the original saved copy again. Changing the snapshot affects upcoming downloads, not previously downloaded CSVs or already published Pinterest Pins.

On Pinterest Exports, use Enrich Pinterest Posts on the relevant import row. The worker combines the Instagram caption, the owning account's brand name, and available analysis of the selected photo. Titles include the brand and the caption's product, collection, or main idea. Descriptions mainly summarize the caption, with natural details supported by the photo analysis.

The copy is saved as enrichment completes. Processing shows an active job; copy-enriched and failed counts show the batch's progress. Enrichment fills missing or outdated copy. Once a batch is fully enriched, Regenerate Copy can explicitly rewrite its saved copy, including after newer image analysis.

Download CSV uses saved current copy where available and caption-based fallback copy for unfinished or outdated items. A download does not trigger a paid AI writing request. Video posts export their saved Instagram video file regardless of import age, and photo posts export a selected photo file. Video Pins use their original imported cover photo (`display_url`) in the Thumbnail column, including the stored server copy of that photo. This is independent of any different carousel image selected for copy analysis. Only videos without a usable cover photo use `00:00` as a first-frame fallback. Photo Pins leave Thumbnail blank. A video post without a saved video file is skipped rather than replaced with a cover image. Each eligible post contributes one selected media asset; it does not automatically become one Pin per carousel slide. Export links use the exact Stored Fashion profile URL for Instagram usernames in the supplied core brand list; other accounts link to the Stored Fashion homepage. Batch exports retain Pinterest tracking parameters. Links update on the next download, including existing batches, without rerunning analysis or copy enrichment. Previously downloaded files and published Pins keep their old links.

Pinterest upload is manual. Mark Exported records the user's export state; it does not publish Pins or confirm a successful Pinterest upload.

What reports and exports contain

Trend reports use a selected niche or supported tagged-brand scope. Standard niche reports use a frozen 14-day Instagram posting window; undated posts are excluded rather than counted by import time. Tagged reports use the existing brand-mention import window and may include creator/customer content. Image analysis runs for the selected source posts before discovery. Large runs produce candidate-evidence parts and a final combined report. Failed parts or incomplete matching responses prevent a completed report from being saved; a successfully checked part with no candidates is allowed.

The AI groups specific visible cues, using an individual saved description for each image. Sparse candidates can combine across parts. Code then checks every eligible image description against registered physical-cue rules, not just the selected gallery examples. All required features on one item must occur together in a supporting phrase. Different prints and materials are kept separate. New image analysis asks for complete item phrases to link garment, color and material evidence. This conservative count can miss attributes that are not explicitly linked in saved descriptions; it does not guess. Titles and taxonomy use the same rules as the counts. The report shows matching posts, distinct posting accounts and the share of eligible posts. Established groups need four supported images; emerging groups need three, across at least two posts and two accounts. At most two images per post contribute to qualification. A photo may support different physical cues, but it is displayed once in the report gallery.

Support score summarizes posting breadth and imported engagement; it is not a probability that the trend is correct. Historical comparisons exclude overlapping reruns, incomplete reports and individual parts. Rising, declining or sustained labels require measured shares from comparable adjacent windows. Otherwise trajectory stays unknown. These observations describe account posting activity, not consumer demand or sales. Missing or irrelevant visual descriptions are excluded and coverage is shown in the report details.

Report overviews attribute accounts only to their verified groups, using available display names. Publishing names, descriptions and captions are edited separately from visual classification. Writing issues are retried once, then neutral verified facts are used so a wording problem does not discard valid evidence. Each trend includes physical cue details, taxonomy fields and search phrases for editorial/product-matching preparation, plus evidence text for gallery images. The Publishing Draft button downloads Markdown with the overview, measured trends and source links. Evidence CSV includes all measured memberships and diagnostic exports retain the saved decisions. Configured RenderForm templates can generate graphics. The app does not automatically publish a weekly article to storedfashion.com. Older incomplete reports show a warning and are not used as comparable history.

Other exports include filtered account/post data, tagged content, prospect lists, website leads, and Pinterest CSVs. Available fields depend on what was imported or enriched.

How to interpret the content

  • An account or user is an Instagram profile. A post belongs to an account. A carousel contains several images; a slide identifies one of those images.
  • A niche tag is an organizational label, such as activewear or beauty. An import batch identifies one import operation. A background-job ID identifies the task that processed it.
  • Captions, account details, and engagement counts come from imported snapshots. They may be incomplete or older than the current Instagram post.
  • Visual labels, profile classifications, trends, and Pinterest copy are AI interpretations. Product names should come from captions or other source evidence; inferred colors, materials, or trends are not independently verified facts.
  • Tagged content may be posted by a creator who mentions a brand. The posting account and the mentioned brand are different concepts.
  • Import totals can include both newly created and updated records. Exportable counts can be lower because media is unavailable or records now belong to a newer import batch.

Implementation context for AI assistants

The app uses Python/Flask for pages, SQLAlchemy with PostgreSQL for saved data, and Redis/RQ for background work. Separate services handle imports, image analysis, profile enrichment, Pinterest copy, reports, and graphics. External integrations include Apify, OpenRouter, Gmail, and RenderForm. This guide describes the app's behavior; it is not an API for accessing private records.